feat(annotate): add plan toggle, drop subtask verify pass, 4xH200 job

- PlanConfig.emit_plan (default True): keep subtasks + memory but skip
  the per-boundary "plan" rows and their VLM call when False.
- Remove the subtask_verify pass entirely: pruning dropped legitimate
  subtasks and the stitch step already guarantees full-episode coverage.
  Deletes _verify_subtasks, both call sites, and the now-unused
  module_1_subtask_verify prompt.
- run_hf_job example: 4xH200 (4 vllm servers), emit_plan=false, vqa off.

Co-authored-by: Cursor <cursoragent@cursor.com>

.github/workflows/benchmark_tests.yml

@@ -382,6 +382,7 @@ jobs:
                 --policy.path=\"\$ROBOTWIN_POLICY\" \
                 --env.type=robotwin \
                 --env.task=\"\$ROBOTWIN_TASKS\" \
+                --env.max_parallel_tasks=5 \
                 --eval.batch_size=1 \
                 --eval.n_episodes=1 \
                 --eval.use_async_envs=false \
@@ -482,6 +483,7 @@ jobs:
                 --policy.path=lerobot/smolvla_robocasa \
                 --env.type=robocasa \
                 --env.task=CloseFridge,OpenCabinet,OpenDrawer,TurnOnMicrowave,TurnOffStove,CloseToasterOvenDoor,SlideDishwasherRack,TurnOnSinkFaucet,NavigateKitchen,TurnOnElectricKettle \
+                --env.max_parallel_tasks=5 \
                 --eval.batch_size=1 \
                 --eval.n_episodes=1 \
                 --eval.use_async_envs=false \
@@ -693,6 +695,7 @@ jobs:
                 --env.task=\"\$ROBOMME_TASKS\" \
                 --env.dataset_split=test \
                 --env.task_ids=[0] \
+                --env.max_parallel_tasks=5 \
                 --eval.batch_size=1 \
                 --eval.n_episodes=1 \
                 --eval.use_async_envs=false \
@@ -800,6 +803,7 @@ jobs:
                 --env.type=libero_plus \
                 --env.task=\"\$LIBERO_PLUS_SUITE\" \
                 --env.task_ids=\"\$LIBERO_PLUS_TASK_IDS\" \
+                --env.max_parallel_tasks=5 \
                 --eval.batch_size=1 \
                 --eval.n_episodes=1 \
                 --eval.use_async_envs=false \
@@ -900,6 +904,8 @@ jobs:
                 --policy.path=lerobot/smolvla_vlabench \
                 --env.type=vlabench \
                 --env.task=select_fruit,select_toy,select_book,select_painting,select_drink,select_ingredient,select_billiards,select_poker,add_condiment,insert_flower \
+                --env.episode_length=50 \
+                --env.max_parallel_tasks=5 \
                 --eval.batch_size=1 \
                 --eval.n_episodes=1 \
                 --eval.use_async_envs=false \

.github/workflows/documentation-upload-pr.yml

@@ -33,7 +33,7 @@ jobs:
       github.event.workflow_run.event == 'pull_request' &&
       github.event.workflow_run.conclusion == 'success' &&
       github.repository == 'huggingface/lerobot'
-    uses: huggingface/doc-builder/.github/workflows/upload_pr_documentation.yml@9ad2de8582b56c017cb530c1165116d40433f1c6  # main
+    uses: huggingface/doc-builder/.github/workflows/upload_pr_documentation.yml@2430c1ec91d04667414e2fa31ecfc36c153ea391  # main
     with:
       package_name: lerobot
     secrets:

.github/workflows/documentation.yml

@@ -55,7 +55,7 @@ jobs:
       github.repository == 'huggingface/lerobot'
     permissions:
       contents: read
-    uses: huggingface/doc-builder/.github/workflows/build_main_documentation.yml@90b4ee2c10b81b5c1a6367c4e6fc9e2fb510a7e3  # main
+    uses: huggingface/doc-builder/.github/workflows/build_main_documentation.yml@2430c1ec91d04667414e2fa31ecfc36c153ea391  # main
     with:
       commit_sha: ${{ github.sha }}
       package: lerobot
@@ -78,7 +78,7 @@ jobs:
     permissions:
       contents: read
       pull-requests: write
-    uses: huggingface/doc-builder/.github/workflows/build_pr_documentation.yml@90b4ee2c10b81b5c1a6367c4e6fc9e2fb510a7e3  # main
+    uses: huggingface/doc-builder/.github/workflows/build_pr_documentation.yml@2430c1ec91d04667414e2fa31ecfc36c153ea391  # main
     with:
       commit_sha: ${{ github.event.pull_request.head.sha }}
       pr_number: ${{ github.event.number }}

.github/workflows/release.yml

@@ -152,13 +152,14 @@ jobs:
             BASE_VERSION="${VERSION%%-*}"
             echo "Installing pre-release version $BASE_VERSION from TestPyPI..."
             uv pip install \
+              --torch-backend cpu \
               --index-url https://test.pypi.org/simple/ \
               --extra-index-url https://pypi.org/simple \
               --index-strategy unsafe-best-match \
                "lerobot[all]==$BASE_VERSION"
           else
             echo "Installing release version $VERSION from PyPI..."
-            uv pip install "lerobot[all]==$VERSION"
+            uv pip install --torch-backend cpu "lerobot[all]==$VERSION"
           fi
       - name: Check lerobot version
         run: uv run python -c "import lerobot; print(lerobot.__version__)"

.github/workflows/stale.yml

@@ -19,19 +19,19 @@ on:
   workflow_dispatch:
 
   # Runs at 02:00
-  schedule:
-    - cron: "0 2 * * *"
+  # schedule:
+  #   - cron: "0 2 * * *"
 
 env:
   CLOSE_ISSUE_MESSAGE: >
-    This issue was closed because it has been stalled for 14 days with no activity.
+    This issue was closed because it has been stalled for 30 days with no activity.
     Feel free to reopen if is still relevant, or to ping a collaborator if you have any questions.
   CLOSE_PR_MESSAGE: >
-    This PR was closed because it has been stalled for 21 days with no activity.
+    This PR was closed because it has been stalled for 30 days with no activity.
     Feel free to reopen if is still relevant, or to ping a collaborator if you have any questions.
   WARN_ISSUE_MESSAGE: >
     This issue has been automatically marked as stale because it has not had
-    recent activity (6 months). It will be closed if no further activity occurs.
+    recent activity (1 year). It will be closed if no further activity occurs.
     Any change, comment or update to this issue will reset this count.
     Thank you for your contributions.
   WARN_PR_MESSAGE: >
@@ -59,10 +59,10 @@ jobs:
           stale-pr-label: stale
           exempt-issue-labels: never-stale
           exempt-pr-labels: never-stale
-          days-before-issue-stale: 180
-          days-before-issue-close: 14
+          days-before-issue-stale: 365
+          days-before-issue-close: 30
           days-before-pr-stale: 365
-          days-before-pr-close: 21
+          days-before-pr-close: 30
           delete-branch: true
           close-issue-message: ${{ env.CLOSE_ISSUE_MESSAGE }}
           close-pr-message: ${{ env.CLOSE_PR_MESSAGE }}

AGENT_GUIDE.md

@@ -232,6 +232,8 @@ Match the policy to the user's **GPU memory** and **time budget**. Numbers below
 
 All policies typically train for **5–10 epochs** (see §7).
 
+> **Human-facing version:** the [Compute Hardware Guide](./docs/source/hardware_guide.mdx) reuses the table below and adds a cloud-GPU tier guide and a Hugging Face Jobs pointer.
+
 | Policy      | Batch | Update (ms) | Peak GPU mem (GB) | Best for                                                                                         |
 | ----------- | ----: | ----------: | ----------------: | ------------------------------------------------------------------------------------------------ |
 | `act`       |     4 |    **83.9** |          **0.94** | First-time users, laptops, single-task. Fast and reliable.                                       |

MANIFEST.in

@@ -1,3 +1,4 @@
 include src/lerobot/templates/lerobot_modelcard_template.md
+include src/lerobot/templates/lerobot_rewardmodel_modelcard_template.md
 include src/lerobot/datasets/card_template.md
 include src/lerobot/envs/metaworld_config.json

Makefile

@@ -178,3 +178,9 @@ test-smolvla-ete-eval:
 		--env.episode_length=5 \
 		--eval.n_episodes=1 \
 		--eval.batch_size=1
+
+# E2E annotation pipeline smoke test against a tiny in-memory fixture
+# dataset. Opt-in (not part of `make test-end-to-end`) and uses a stub VLM
+# backend, so it does not require a real model checkpoint or GPU.
+annotation-e2e:
+	uv run python -m tests.annotations.run_e2e_smoke

README.md

@@ -109,7 +109,7 @@ lerobot-train \
 
 Similarly to the hardware, you can easily implement your own policy & leverage LeRobot's data collection, training, and visualization tools, and share your model to the HF Hub
 
-For detailed policy setup guides, see the [Policy Documentation](https://huggingface.co/docs/lerobot/bring_your_own_policies).
+For detailed policy setup guides, see the [Policy Documentation](https://huggingface.co/docs/lerobot/bring_your_own_policies). For GPU/RAM requirements and expected training time per policy, see the [Compute Hardware Guide](https://huggingface.co/docs/lerobot/hardware_guide).
 
 ## Inference & Evaluation
 

benchmarks/video/README.md

@@ -1,288 +0,0 @@
-# Video benchmark
-
-## Questions
-
-What is the optimal trade-off between:
-
-- maximizing loading time with random access,
-- minimizing memory space on disk,
-- maximizing success rate of policies,
-- compatibility across devices/platforms for decoding videos (e.g. video players, web browsers).
-
-How to encode videos?
-
-- Which video codec (`-vcodec`) to use? h264, h265, AV1?
-- What pixel format to use (`-pix_fmt`)? `yuv444p` or `yuv420p`?
-- How much compression (`-crf`)? No compression with `0`, intermediate compression with `25` or extreme with `50+`?
-- Which frequency to chose for key frames (`-g`)? A key frame every `10` frames?
-
-How to decode videos?
-
-- Which `decoder`? `torchvision`, `torchaudio`, `ffmpegio`, `decord`, or `nvc`?
-- What scenarios to use for the requesting timestamps during benchmark? (`timestamps_mode`)
-
-## Variables
-
-**Image content & size**
-We don't expect the same optimal settings for a dataset of images from a simulation, or from real-world in an apartment, or in a factory, or outdoor, or with lots of moving objects in the scene, etc. Similarly, loading times might not vary linearly with the image size (resolution).
-For these reasons, we run this benchmark on four representative datasets:
-
-- `lerobot/pusht_image`: (96 x 96 pixels) simulation with simple geometric shapes, fixed camera.
-- `lerobot/aloha_mobile_shrimp_image`: (480 x 640 pixels) real-world indoor, moving camera.
-- `lerobot/paris_street`: (720 x 1280 pixels) real-world outdoor, moving camera.
-- `lerobot/kitchen`: (1080 x 1920 pixels) real-world indoor, fixed camera.
-
-Note: The datasets used for this benchmark need to be image datasets, not video datasets.
-
-**Data augmentations**
-We might revisit this benchmark and find better settings if we train our policies with various data augmentations to make them more robust (e.g. robust to color changes, compression, etc.).
-
-### Encoding parameters
-
-| parameter   | values                                                       |
-| ----------- | ------------------------------------------------------------ |
-| **vcodec**  | `libx264`, `libx265`, `libsvtav1`                            |
-| **pix_fmt** | `yuv444p`, `yuv420p`                                         |
-| **g**       | `1`, `2`, `3`, `4`, `5`, `6`, `10`, `15`, `20`, `40`, `None` |
-| **crf**     | `0`, `5`, `10`, `15`, `20`, `25`, `30`, `40`, `50`, `None`   |
-
-Note that `crf` value might be interpreted differently by various video codecs. In other words, the same value used with one codec doesn't necessarily translate into the same compression level with another codec. In fact, the default value (`None`) isn't the same amongst the different video codecs. Importantly, it is also the case for many other ffmpeg arguments like `g` which specifies the frequency of the key frames.
-
-For a comprehensive list and documentation of these parameters, see the ffmpeg documentation depending on the video codec used:
-
-- h264: https://trac.ffmpeg.org/wiki/Encode/H.264
-- h265: https://trac.ffmpeg.org/wiki/Encode/H.265
-- AV1: https://trac.ffmpeg.org/wiki/Encode/AV1
-
-### Decoding parameters
-
-**Decoder**
-We tested two video decoding backends from torchvision:
-
-- `pyav`
-- `video_reader` (requires to build torchvision from source)
-
-**Requested timestamps**
-Given the way video decoding works, once a keyframe has been loaded, the decoding of subsequent frames is fast.
-This of course is affected by the `-g` parameter during encoding, which specifies the frequency of the keyframes. Given our typical use cases in robotics policies which might request a few timestamps in different random places, we want to replicate these use cases with the following scenarios:
-
-- `1_frame`: 1 frame,
-- `2_frames`: 2 consecutive frames (e.g. `[t, t + 1 / fps]`),
-- `6_frames`: 6 consecutive frames (e.g. `[t + i / fps for i in range(6)]`)
-
-Note that this differs significantly from a typical use case like watching a movie, in which every frame is loaded sequentially from the beginning to the end and it's acceptable to have big values for `-g`.
-
-Additionally, because some policies might request single timestamps that are a few frames apart, we also have the following scenario:
-
-- `2_frames_4_space`: 2 frames with 4 consecutive frames of spacing in between (e.g `[t, t + 5 / fps]`),
-
-However, due to how video decoding is implemented with `pyav`, we don't have access to an accurate seek so in practice this scenario is essentially the same as `6_frames` since all 6 frames between `t` and `t + 5 / fps` will be decoded.
-
-## Metrics
-
-**Data compression ratio (lower is better)**
-`video_images_size_ratio` is the ratio of the memory space on disk taken by the encoded video over the memory space taken by the original images. For instance, `video_images_size_ratio=25%` means that the video takes 4 times less memory space on disk compared to the original images.
-
-**Loading time ratio (lower is better)**
-`video_images_load_time_ratio` is the ratio of the time it takes to decode frames from the video at a given timestamps over the time it takes to load the exact same original images. Lower is better. For instance, `video_images_load_time_ratio=200%` means that decoding from video is 2 times slower than loading the original images.
-
-**Average Mean Square Error (lower is better)**
-`avg_mse` is the average mean square error between each decoded frame and its corresponding original image over all requested timestamps, and also divided by the number of pixels in the image to be comparable when switching to different image sizes.
-
-**Average Peak Signal to Noise Ratio (higher is better)**
-`avg_psnr` measures the ratio between the maximum possible power of a signal and the power of corrupting noise that affects the fidelity of its representation. Higher PSNR indicates better quality.
-
-**Average Structural Similarity Index Measure (higher is better)**
-`avg_ssim` evaluates the perceived quality of images by comparing luminance, contrast, and structure. SSIM values range from -1 to 1, where 1 indicates perfect similarity.
-
-One aspect that can't be measured here with those metrics is the compatibility of the encoding across platforms, in particular on web browser, for visualization purposes.
-h264, h265 and AV1 are all commonly used codecs and should not pose an issue. However, the chroma subsampling (`pix_fmt`) format might affect compatibility:
-
-- `yuv420p` is more widely supported across various platforms, including web browsers.
-- `yuv444p` offers higher color fidelity but might not be supported as broadly.
-
-<!-- **Loss of a pretrained policy (higher is better)** (not available)
-`loss_pretrained` is the result of evaluating with the selected encoding/decoding settings a policy pretrained on original images. It is easier to understand than `avg_l2_error`.
-
-**Success rate after retraining (higher is better)** (not available)
-`success_rate` is the result of training and evaluating a policy with the selected encoding/decoding settings. It is the most difficult metric to get but also the very best. -->
-
-## How the benchmark works
-
-The benchmark evaluates both encoding and decoding of video frames on the first episode of each dataset.
-
-**Encoding:** for each `vcodec` and `pix_fmt` pair, we use a default value for `g` and `crf` upon which we change a single value (either `g` or `crf`) to one of the specified values (we don't test every combination of those as this would be computationally too heavy).
-This gives a unique set of encoding parameters which is used to encode the episode.
-
-**Decoding:** Then, for each of those unique encodings, we iterate through every combination of the decoding parameters `backend` and `timestamps_mode`. For each of them, we record the metrics of a number of samples (given by `--num-samples`). This is parallelized for efficiency and the number of processes can be controlled with `--num-workers`. Ideally, it's best to have a `--num-samples` that is divisible by `--num-workers`.
-
-Intermediate results saved for each `vcodec` and `pix_fmt` combination in csv tables.
-These are then all concatenated to a single table ready for analysis.
-
-## Caveats
-
-We tried to measure the most impactful parameters for both encoding and decoding. However, for computational reasons we can't test out every combination.
-
-Additional encoding parameters exist that are not included in this benchmark. In particular:
-
-- `-preset` which allows for selecting encoding presets. This represents a collection of options that will provide a certain encoding speed to compression ratio. By leaving this parameter unspecified, it is considered to be `medium` for libx264 and libx265 and `8` for libsvtav1.
-- `-tune` which allows to optimize the encoding for certain aspects (e.g. film quality, fast decoding, etc.).
-
-See the documentation mentioned above for more detailed info on these settings and for a more comprehensive list of other parameters.
-
-Similarly on the decoding side, other decoders exist but are not implemented in our current benchmark. To name a few:
-
-- `torchaudio`
-- `ffmpegio`
-- `decord`
-- `nvc`
-
-Note as well that since we are mostly interested in the performance at decoding time (also because encoding is done only once before uploading a dataset), we did not measure encoding times nor have any metrics regarding encoding.
-However, besides the necessity to build ffmpeg from source, encoding did not pose any issue and it didn't take a significant amount of time during this benchmark.
-
-## Install
-
-Building ffmpeg from source is required to include libx265 and libaom/libsvtav1 (av1) video codecs ([compilation guide](https://trac.ffmpeg.org/wiki/CompilationGuide/Ubuntu)).
-
-**Note:** While you still need to build torchvision with a conda-installed `ffmpeg<4.3` to use the `video_reader` decoder (as described in [#220](https://github.com/huggingface/lerobot/pull/220)), you also need another version which is custom-built with all the video codecs for encoding. For the script to then use that version, you can prepend the command above with `PATH="$HOME/bin:$PATH"`, which is where ffmpeg should be built.
-
-## Adding a video decoder
-
-Right now, we're only benchmarking the two video decoder available with torchvision: `pyav` and `video_reader`.
-You can easily add a new decoder to benchmark by adding it to this function in the script:
-
-```diff
-def decode_video_frames(
-    video_path: str,
-    timestamps: list[float],
-    tolerance_s: float,
-    backend: str,
-) -> torch.Tensor:
-    if backend in ["pyav", "video_reader"]:
-        return decode_video_frames_torchvision(
-            video_path, timestamps, tolerance_s, backend
-        )
-+    elif backend == ["your_decoder"]:
-+        return your_decoder_function(
-+            video_path, timestamps, tolerance_s, backend
-+        )
-    else:
-        raise NotImplementedError(backend)
-```
-
-## Example
-
-For a quick run, you can try these parameters:
-
-```bash
-python benchmark/video/run_video_benchmark.py \
-    --output-dir outputs/video_benchmark \
-    --repo-ids \
-        lerobot/pusht_image \
-        lerobot/aloha_mobile_shrimp_image \
-    --vcodec libx264 libx265 \
-    --pix-fmt yuv444p yuv420p \
-    --g 2 20 None \
-    --crf 10 40 None \
-    --timestamps-modes 1_frame 2_frames \
-    --backends pyav video_reader \
-    --num-samples 5 \
-    --num-workers 5 \
-    --save-frames 0
-```
-
-## Results
-
-### Reproduce
-
-We ran the benchmark with the following parameters:
-
-```bash
-# h264 and h265 encodings
-python benchmark/video/run_video_benchmark.py \
-    --output-dir outputs/video_benchmark \
-    --repo-ids \
-        lerobot/pusht_image \
-        lerobot/aloha_mobile_shrimp_image \
-        lerobot/paris_street \
-        lerobot/kitchen \
-    --vcodec libx264 libx265 \
-    --pix-fmt yuv444p yuv420p \
-    --g 1 2 3 4 5 6 10 15 20 40 None \
-    --crf 0 5 10 15 20 25 30 40 50 None \
-    --timestamps-modes 1_frame 2_frames 6_frames \
-    --backends pyav video_reader \
-    --num-samples 50 \
-    --num-workers 5 \
-    --save-frames 1
-
-# av1 encoding (only compatible with yuv420p and pyav decoder)
-python benchmark/video/run_video_benchmark.py \
-    --output-dir outputs/video_benchmark \
-    --repo-ids \
-        lerobot/pusht_image \
-        lerobot/aloha_mobile_shrimp_image \
-        lerobot/paris_street \
-        lerobot/kitchen \
-    --vcodec libsvtav1 \
-    --pix-fmt yuv420p \
-    --g 1 2 3 4 5 6 10 15 20 40 None \
-    --crf 0 5 10 15 20 25 30 40 50 None \
-    --timestamps-modes 1_frame 2_frames 6_frames \
-    --backends pyav \
-    --num-samples 50 \
-    --num-workers 5 \
-    --save-frames 1
-```
-
-The full results are available [here](https://docs.google.com/spreadsheets/d/1OYJB43Qu8fC26k_OyoMFgGBBKfQRCi4BIuYitQnq3sw/edit?usp=sharing)
-
-### Parameters selected for LeRobotDataset
-
-Considering these results, we chose what we think is the best set of encoding parameter:
-
-- vcodec: `libsvtav1`
-- pix-fmt: `yuv420p`
-- g: `2`
-- crf: `30`
-
-Since we're using av1 encoding, we're choosing the `pyav` decoder as `video_reader` does not support it (and `pyav` doesn't require a custom build of `torchvision`).
-
-### Summary
-
-These tables show the results for `g=2` and `crf=30`, using `timestamps-modes=6_frames` and `backend=pyav`
-
-| video_images_size_ratio           | vcodec     | pix_fmt |           |           |           |
-| --------------------------------- | ---------- | ------- | --------- | --------- | --------- |
-|                                   | libx264    |         | libx265   |           | libsvtav1 |
-| repo_id                           | yuv420p    | yuv444p | yuv420p   | yuv444p   | yuv420p   |
-| lerobot/pusht_image               | **16.97%** | 17.58%  | 18.57%    | 18.86%    | 22.06%    |
-| lerobot/aloha_mobile_shrimp_image | 2.14%      | 2.11%   | 1.38%     | **1.37%** | 5.59%     |
-| lerobot/paris_street              | 2.12%      | 2.13%   | **1.54%** | **1.54%** | 4.43%     |
-| lerobot/kitchen                   | 1.40%      | 1.39%   | **1.00%** | **1.00%** | 2.52%     |
-
-| video_images_load_time_ratio      | vcodec  | pix_fmt |          |         |           |
-| --------------------------------- | ------- | ------- | -------- | ------- | --------- |
-|                                   | libx264 |         | libx265  |         | libsvtav1 |
-| repo_id                           | yuv420p | yuv444p | yuv420p  | yuv444p | yuv420p   |
-| lerobot/pusht_image               | 6.45    | 5.19    | **1.90** | 2.12    | 2.47      |
-| lerobot/aloha_mobile_shrimp_image | 11.80   | 7.92    | 0.71     | 0.85    | **0.48**  |
-| lerobot/paris_street              | 2.21    | 2.05    | 0.36     | 0.49    | **0.30**  |
-| lerobot/kitchen                   | 1.46    | 1.46    | 0.28     | 0.51    | **0.26**  |
-
-|                                   |          | vcodec   | pix_fmt      |          |           |              |
-| --------------------------------- | -------- | -------- | ------------ | -------- | --------- | ------------ |
-|                                   |          | libx264  |              | libx265  |           | libsvtav1    |
-| repo_id                           | metric   | yuv420p  | yuv444p      | yuv420p  | yuv444p   | yuv420p      |
-| lerobot/pusht_image               | avg_mse  | 2.90E-04 | **2.03E-04** | 3.13E-04 | 2.29E-04  | 2.19E-04     |
-|                                   | avg_psnr | 35.44    | 37.07        | 35.49    | **37.30** | 37.20        |
-|                                   | avg_ssim | 98.28%   | **98.85%**   | 98.31%   | 98.84%    | 98.72%       |
-| lerobot/aloha_mobile_shrimp_image | avg_mse  | 2.76E-04 | 2.59E-04     | 3.17E-04 | 3.06E-04  | **1.30E-04** |
-|                                   | avg_psnr | 35.91    | 36.21        | 35.88    | 36.09     | **40.17**    |
-|                                   | avg_ssim | 95.19%   | 95.18%       | 95.00%   | 95.05%    | **97.73%**   |
-| lerobot/paris_street              | avg_mse  | 6.89E-04 | 6.70E-04     | 4.03E-03 | 4.02E-03  | **3.09E-04** |
-|                                   | avg_psnr | 33.48    | 33.68        | 32.05    | 32.15     | **35.40**    |
-|                                   | avg_ssim | 93.76%   | 93.75%       | 89.46%   | 89.46%    | **95.46%**   |
-| lerobot/kitchen                   | avg_mse  | 2.50E-04 | 2.24E-04     | 4.28E-04 | 4.18E-04  | **1.53E-04** |
-|                                   | avg_psnr | 36.73    | 37.33        | 36.56    | 36.75     | **39.12**    |
-|                                   | avg_ssim | 95.47%   | 95.58%       | 95.52%   | 95.53%    | **96.82%**   |

benchmarks/video/run_video_benchmark.py

@@ -1,488 +0,0 @@
-#!/usr/bin/env python
-
-# Copyright 2024 The HuggingFace Inc. team. All rights reserved.
-#
-# Licensed under the Apache License, Version 2.0 (the "License");
-# you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at
-#
-#     http://www.apache.org/licenses/LICENSE-2.0
-#
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-# limitations under the License.
-"""Assess the performance of video decoding in various configurations.
-
-This script will benchmark different video encoding and decoding parameters.
-See the provided README.md or run `python benchmark/video/run_video_benchmark.py --help` for usage info.
-"""
-
-import argparse
-import datetime as dt
-import itertools
-import random
-import shutil
-from collections import OrderedDict
-from concurrent.futures import ThreadPoolExecutor, as_completed
-from pathlib import Path
-from threading import Lock
-
-import einops
-import numpy as np
-import pandas as pd
-import PIL
-import torch
-from skimage.metrics import mean_squared_error, peak_signal_noise_ratio, structural_similarity
-from tqdm import tqdm
-
-from lerobot.datasets.lerobot_dataset import LeRobotDataset
-from lerobot.datasets.video_utils import (
-    decode_video_frames,
-    encode_video_frames,
-)
-from lerobot.utils.constants import OBS_IMAGE
-from lerobot.utils.utils import TimerManager
-
-BASE_ENCODING = OrderedDict(
-    [
-        ("vcodec", "libx264"),
-        ("pix_fmt", "yuv444p"),
-        ("g", 2),
-        ("crf", None),
-        # TODO(aliberts): Add fastdecode
-        # ("fastdecode", 0),
-    ]
-)
-
-
-# TODO(rcadene, aliberts): move to `utils.py` folder when we want to refactor
-def parse_int_or_none(value) -> int | None:
-    if value.lower() == "none":
-        return None
-    try:
-        return int(value)
-    except ValueError as e:
-        raise argparse.ArgumentTypeError(f"Invalid int or None: {value}") from e
-
-
-def check_datasets_formats(repo_ids: list) -> None:
-    for repo_id in repo_ids:
-        dataset = LeRobotDataset(repo_id)
-        if len(dataset.meta.video_keys) > 0:
-            raise ValueError(
-                f"Use only image dataset for running this benchmark. Video dataset provided: {repo_id}"
-            )
-
-
-def get_directory_size(directory: Path) -> int:
-    total_size = 0
-    for item in directory.rglob("*"):
-        if item.is_file():
-            total_size += item.stat().st_size
-    return total_size
-
-
-def load_original_frames(imgs_dir: Path, timestamps: list[float], fps: int) -> torch.Tensor:
-    frames = []
-    for ts in timestamps:
-        idx = int(ts * fps)
-        frame = PIL.Image.open(imgs_dir / f"frame-{idx:06d}.png")
-        frame = torch.from_numpy(np.array(frame))
-        frame = frame.type(torch.float32) / 255
-        frame = einops.rearrange(frame, "h w c -> c h w")
-        frames.append(frame)
-    return torch.stack(frames)
-
-
-def save_decoded_frames(
-    imgs_dir: Path, save_dir: Path, frames: torch.Tensor, timestamps: list[float], fps: int
-) -> None:
-    if save_dir.exists() and len(list(save_dir.glob("frame-*.png"))) == len(timestamps):
-        return
-
-    save_dir.mkdir(parents=True, exist_ok=True)
-    for i, ts in enumerate(timestamps):
-        idx = int(ts * fps)
-        frame_hwc = (frames[i].permute((1, 2, 0)) * 255).type(torch.uint8).cpu().numpy()
-        PIL.Image.fromarray(frame_hwc).save(save_dir / f"frame-{idx:06d}_decoded.png")
-        shutil.copyfile(imgs_dir / f"frame-{idx:06d}.png", save_dir / f"frame-{idx:06d}_original.png")
-
-
-def save_first_episode(imgs_dir: Path, dataset: LeRobotDataset) -> None:
-    episode_index = 0
-    ep_num_images = dataset.meta.episodes["length"][episode_index]
-    if imgs_dir.exists() and len(list(imgs_dir.glob("frame-*.png"))) == ep_num_images:
-        return
-
-    imgs_dir.mkdir(parents=True, exist_ok=True)
-    hf_dataset = dataset.hf_dataset.with_format(None)
-
-    # We only save images from the first camera
-    img_keys = [key for key in hf_dataset.features if key.startswith(OBS_IMAGE)]
-    imgs_dataset = hf_dataset.select_columns(img_keys[0])
-
-    for i, item in enumerate(
-        tqdm(imgs_dataset, desc=f"saving {dataset.repo_id} first episode images", leave=False)
-    ):
-        img = item[img_keys[0]]
-        img.save(str(imgs_dir / f"frame-{i:06d}.png"), quality=100)
-
-        if i >= ep_num_images - 1:
-            break
-
-
-def sample_timestamps(timestamps_mode: str, ep_num_images: int, fps: int) -> list[float]:
-    # Start at 5 to allow for 2_frames_4_space and 6_frames
-    idx = random.randint(5, ep_num_images - 1)
-    match timestamps_mode:
-        case "1_frame":
-            frame_indexes = [idx]
-        case "2_frames":
-            frame_indexes = [idx - 1, idx]
-        case "2_frames_4_space":
-            frame_indexes = [idx - 5, idx]
-        case "6_frames":
-            frame_indexes = [idx - i for i in range(6)][::-1]
-        case _:
-            raise ValueError(timestamps_mode)
-
-    return [idx / fps for idx in frame_indexes]
-
-
-def benchmark_decoding(
-    imgs_dir: Path,
-    video_path: Path,
-    timestamps_mode: str,
-    backend: str,
-    ep_num_images: int,
-    fps: int,
-    num_samples: int = 50,
-    num_workers: int = 4,
-    save_frames: bool = False,
-) -> dict:
-    def process_sample(sample: int, lock: Lock):
-        time_benchmark = TimerManager(log=False)
-        timestamps = sample_timestamps(timestamps_mode, ep_num_images, fps)
-        num_frames = len(timestamps)
-        result = {
-            "psnr_values": [],
-            "ssim_values": [],
-            "mse_values": [],
-        }
-
-        with time_benchmark, lock:
-            frames = decode_video_frames(video_path, timestamps=timestamps, tolerance_s=5e-1, backend=backend)
-        result["load_time_video_ms"] = (time_benchmark.last * 1000) / num_frames
-
-        with time_benchmark:
-            original_frames = load_original_frames(imgs_dir, timestamps, fps)
-        result["load_time_images_ms"] = (time_benchmark.last * 1000) / num_frames
-
-        frames_np, original_frames_np = frames.numpy(), original_frames.numpy()
-        for i in range(num_frames):
-            result["mse_values"].append(mean_squared_error(original_frames_np[i], frames_np[i]))
-            result["psnr_values"].append(
-                peak_signal_noise_ratio(original_frames_np[i], frames_np[i], data_range=1.0)
-            )
-            result["ssim_values"].append(
-                structural_similarity(original_frames_np[i], frames_np[i], data_range=1.0, channel_axis=0)
-            )
-
-        if save_frames and sample == 0:
-            save_dir = video_path.with_suffix("") / f"{timestamps_mode}_{backend}"
-            save_decoded_frames(imgs_dir, save_dir, frames, timestamps, fps)
-
-        return result
-
-    load_times_video_ms = []
-    load_times_images_ms = []
-    mse_values = []
-    psnr_values = []
-    ssim_values = []
-
-    # A sample is a single set of decoded frames specified by timestamps_mode (e.g. a single frame, 2 frames, etc.).
-    # For each sample, we record metrics (loading time and quality metrics) which are then averaged over all samples.
-    # As these samples are independent, we run them in parallel threads to speed up the benchmark.
-    # Use a single shared lock for all worker threads
-    shared_lock = Lock()
-    with ThreadPoolExecutor(max_workers=num_workers) as executor:
-        futures = [executor.submit(process_sample, i, shared_lock) for i in range(num_samples)]
-        for future in tqdm(as_completed(futures), total=num_samples, desc="samples", leave=False):
-            result = future.result()
-            load_times_video_ms.append(result["load_time_video_ms"])
-            load_times_images_ms.append(result["load_time_images_ms"])
-            psnr_values.extend(result["psnr_values"])
-            ssim_values.extend(result["ssim_values"])
-            mse_values.extend(result["mse_values"])
-
-    avg_load_time_video_ms = float(np.array(load_times_video_ms).mean())
-    avg_load_time_images_ms = float(np.array(load_times_images_ms).mean())
-    video_images_load_time_ratio = avg_load_time_video_ms / avg_load_time_images_ms
-
-    return {
-        "avg_load_time_video_ms": avg_load_time_video_ms,
-        "avg_load_time_images_ms": avg_load_time_images_ms,
-        "video_images_load_time_ratio": video_images_load_time_ratio,
-        "avg_mse": float(np.mean(mse_values)),
-        "avg_psnr": float(np.mean(psnr_values)),
-        "avg_ssim": float(np.mean(ssim_values)),
-    }
-
-
-def benchmark_encoding_decoding(
-    dataset: LeRobotDataset,
-    video_path: Path,
-    imgs_dir: Path,
-    encoding_cfg: dict,
-    decoding_cfg: dict,
-    num_samples: int,
-    num_workers: int,
-    save_frames: bool,
-    overwrite: bool = False,
-    seed: int = 1337,
-) -> list[dict]:
-    fps = dataset.fps
-
-    if overwrite or not video_path.is_file():
-        tqdm.write(f"encoding {video_path}")
-        encode_video_frames(
-            imgs_dir=imgs_dir,
-            video_path=video_path,
-            fps=fps,
-            vcodec=encoding_cfg["vcodec"],
-            pix_fmt=encoding_cfg["pix_fmt"],
-            g=encoding_cfg.get("g"),
-            crf=encoding_cfg.get("crf"),
-            # fast_decode=encoding_cfg.get("fastdecode"),
-            overwrite=True,
-        )
-
-    episode_index = 0
-    ep_num_images = dataset.meta.episodes["length"][episode_index]
-    width, height = tuple(dataset[0][dataset.meta.camera_keys[0]].shape[-2:])
-    num_pixels = width * height
-    video_size_bytes = video_path.stat().st_size
-    images_size_bytes = get_directory_size(imgs_dir)
-    video_images_size_ratio = video_size_bytes / images_size_bytes
-
-    random.seed(seed)
-    benchmark_table = []
-    for timestamps_mode in tqdm(
-        decoding_cfg["timestamps_modes"], desc="decodings (timestamps_modes)", leave=False
-    ):
-        for backend in tqdm(decoding_cfg["backends"], desc="decodings (backends)", leave=False):
-            benchmark_row = benchmark_decoding(
-                imgs_dir,
-                video_path,
-                timestamps_mode,
-                backend,
-                ep_num_images,
-                fps,
-                num_samples,
-                num_workers,
-                save_frames,
-            )
-            benchmark_row.update(
-                **{
-                    "repo_id": dataset.repo_id,
-                    "resolution": f"{width} x {height}",
-                    "num_pixels": num_pixels,
-                    "video_size_bytes": video_size_bytes,
-                    "images_size_bytes": images_size_bytes,
-                    "video_images_size_ratio": video_images_size_ratio,
-                    "timestamps_mode": timestamps_mode,
-                    "backend": backend,
-                },
-                **encoding_cfg,
-            )
-            benchmark_table.append(benchmark_row)
-
-    return benchmark_table
-
-
-def main(
-    output_dir: Path,
-    repo_ids: list[str],
-    vcodec: list[str],
-    pix_fmt: list[str],
-    g: list[int],
-    crf: list[int],
-    # fastdecode: list[int],
-    timestamps_modes: list[str],
-    backends: list[str],
-    num_samples: int,
-    num_workers: int,
-    save_frames: bool,
-):
-    check_datasets_formats(repo_ids)
-    encoding_benchmarks = {
-        "g": g,
-        "crf": crf,
-        # "fastdecode": fastdecode,
-    }
-    decoding_benchmarks = {
-        "timestamps_modes": timestamps_modes,
-        "backends": backends,
-    }
-    headers = ["repo_id", "resolution", "num_pixels"]
-    headers += list(BASE_ENCODING.keys())
-    headers += [
-        "timestamps_mode",
-        "backend",
-        "video_size_bytes",
-        "images_size_bytes",
-        "video_images_size_ratio",
-        "avg_load_time_video_ms",
-        "avg_load_time_images_ms",
-        "video_images_load_time_ratio",
-        "avg_mse",
-        "avg_psnr",
-        "avg_ssim",
-    ]
-    file_paths = []
-    for video_codec in tqdm(vcodec, desc="encodings (vcodec)"):
-        for pixel_format in tqdm(pix_fmt, desc="encodings (pix_fmt)", leave=False):
-            benchmark_table = []
-            for repo_id in tqdm(repo_ids, desc="encodings (datasets)", leave=False):
-                dataset = LeRobotDataset(repo_id)
-                imgs_dir = output_dir / "images" / dataset.repo_id.replace("/", "_")
-                # We only use the first episode
-                save_first_episode(imgs_dir, dataset)
-                for duet in [
-                    dict(zip(encoding_benchmarks.keys(), unique_combination, strict=False))
-                    for unique_combination in itertools.product(*encoding_benchmarks.values())
-                ]:
-                    encoding_cfg = BASE_ENCODING.copy()
-                    encoding_cfg["vcodec"] = video_codec
-                    encoding_cfg["pix_fmt"] = pixel_format
-                    for key, value in duet.items():
-                        encoding_cfg[key] = value
-                    args_path = Path("_".join(str(value) for value in encoding_cfg.values()))
-                    video_path = output_dir / "videos" / args_path / f"{repo_id.replace('/', '_')}.mp4"
-                    benchmark_table += benchmark_encoding_decoding(
-                        dataset,
-                        video_path,
-                        imgs_dir,
-                        encoding_cfg,
-                        decoding_benchmarks,
-                        num_samples,
-                        num_workers,
-                        save_frames,
-                    )
-
-            # Save intermediate results
-            benchmark_df = pd.DataFrame(benchmark_table, columns=headers)
-            now = dt.datetime.now()
-            csv_path = (
-                output_dir
-                / f"{now:%Y-%m-%d}_{now:%H-%M-%S}_{video_codec}_{pixel_format}_{num_samples}-samples.csv"
-            )
-            benchmark_df.to_csv(csv_path, header=True, index=False)
-            file_paths.append(csv_path)
-            del benchmark_df
-
-    # Concatenate all results
-    df_list = [pd.read_csv(csv_path) for csv_path in file_paths]
-    concatenated_df = pd.concat(df_list, ignore_index=True)
-    concatenated_path = output_dir / f"{now:%Y-%m-%d}_{now:%H-%M-%S}_all_{num_samples}-samples.csv"
-    concatenated_df.to_csv(concatenated_path, header=True, index=False)
-
-
-if __name__ == "__main__":
-    parser = argparse.ArgumentParser()
-    parser.add_argument(
-        "--output-dir",
-        type=Path,
-        default=Path("outputs/video_benchmark"),
-        help="Directory where the video benchmark outputs are written.",
-    )
-    parser.add_argument(
-        "--repo-ids",
-        type=str,
-        nargs="*",
-        default=[
-            "lerobot/pusht_image",
-            "lerobot/aloha_mobile_shrimp_image",
-            "lerobot/paris_street",
-            "lerobot/kitchen",
-        ],
-        help="Datasets repo-ids to test against. First episodes only are used. Must be images.",
-    )
-    parser.add_argument(
-        "--vcodec",
-        type=str,
-        nargs="*",
-        default=["h264", "hevc", "libsvtav1"],
-        help="Video codecs to be tested",
-    )
-    parser.add_argument(
-        "--pix-fmt",
-        type=str,
-        nargs="*",
-        default=["yuv444p", "yuv420p"],
-        help="Pixel formats (chroma subsampling) to be tested",
-    )
-    parser.add_argument(
-        "--g",
-        type=parse_int_or_none,
-        nargs="*",
-        default=[1, 2, 3, 4, 5, 6, 10, 15, 20, 40, 100, None],
-        help="Group of pictures sizes to be tested.",
-    )
-    parser.add_argument(
-        "--crf",
-        type=parse_int_or_none,
-        nargs="*",
-        default=[0, 5, 10, 15, 20, 25, 30, 40, 50, None],
-        help="Constant rate factors to be tested.",
-    )
-    # parser.add_argument(
-    #     "--fastdecode",
-    #     type=int,
-    #     nargs="*",
-    #     default=[0, 1],
-    #     help="Use the fastdecode tuning option. 0 disables it. "
-    #         "For libx264 and libx265/hevc, only 1 is possible. "
-    #         "For libsvtav1, 1, 2 or 3 are possible values with a higher number meaning a faster decoding optimization",
-    # )
-    parser.add_argument(
-        "--timestamps-modes",
-        type=str,
-        nargs="*",
-        default=[
-            "1_frame",
-            "2_frames",
-            "2_frames_4_space",
-            "6_frames",
-        ],
-        help="Timestamps scenarios to be tested.",
-    )
-    parser.add_argument(
-        "--backends",
-        type=str,
-        nargs="*",
-        default=["torchcodec", "pyav"],
-        help="Torchvision decoding backend to be tested.",
-    )
-    parser.add_argument(
-        "--num-samples",
-        type=int,
-        default=50,
-        help="Number of samples for each encoding x decoding config.",
-    )
-    parser.add_argument(
-        "--num-workers",
-        type=int,
-        default=10,
-        help="Number of processes for parallelized sample processing.",
-    )
-    parser.add_argument(
-        "--save-frames",
-        type=int,
-        default=0,
-        help="Whether to save decoded frames or not. Enter a non-zero number for true.",
-    )
-    args = parser.parse_args()
-    main(**vars(args))

docker/Dockerfile.benchmark.robotwin

@@ -35,7 +35,7 @@ USER root
 ARG ROBOTWIN_SHA=0aeea2d669c0f8516f4d5785f0aa33ba812c14b4
 RUN apt-get update \
     && apt-get install -y --no-install-recommends \
-         cuda-nvcc-12-4 cuda-cudart-dev-12-4 \
+         cuda-nvcc-12-8 cuda-cudart-dev-12-8 \
          libvulkan1 vulkan-tools \
     && mkdir -p /usr/share/vulkan/icd.d \
     && echo '{"file_format_version":"1.0.0","ICD":{"library_path":"libGLX_nvidia.so.0","api_version":"1.3.0"}}' \

docker/Dockerfile.internal

@@ -18,9 +18,8 @@
 # docker build -f docker/Dockerfile.internal -t lerobot-internal .
 
 # Configure the base image for CI with GPU access
-# TODO(Steven): Bump these versions
-ARG CUDA_VERSION=12.4.1
-ARG OS_VERSION=22.04
+ARG CUDA_VERSION=12.8.1
+ARG OS_VERSION=24.04
 FROM nvidia/cuda:${CUDA_VERSION}-base-ubuntu${OS_VERSION}
 
 # Define Python version argument
@@ -36,16 +35,13 @@ ENV DEBIAN_FRONTEND=noninteractive \
 
 # Install Python, system dependencies, and uv (as root)
 RUN apt-get update && apt-get install -y --no-install-recommends \
-    software-properties-common build-essential git curl \
-    libglib2.0-0 libgl1-mesa-glx libegl1-mesa ffmpeg \
+    build-essential git curl \
+    libglib2.0-0 libgl1 libegl1 ffmpeg \
     libusb-1.0-0-dev speech-dispatcher libgeos-dev portaudio19-dev \
     cmake pkg-config ninja-build \
-    && add-apt-repository -y ppa:deadsnakes/ppa \
-    && apt-get update \
-    && apt-get install -y --no-install-recommends \
-       python${PYTHON_VERSION} \
-       python${PYTHON_VERSION}-venv \
-       python${PYTHON_VERSION}-dev \
+    python${PYTHON_VERSION} \
+    python${PYTHON_VERSION}-venv \
+    python${PYTHON_VERSION}-dev \
     && curl -LsSf https://astral.sh/uv/install.sh | sh \
     && mv /root/.local/bin/uv /usr/local/bin/uv \
     && useradd --create-home --shell /bin/bash user_lerobot \

docs/source/_toctree.yml

@@ -3,12 +3,14 @@
     title: LeRobot
   - local: installation
     title: Installation
+  - local: cheat-sheet
+    title: Cheat sheet
   title: Get started
 - sections:
   - local: il_robots
     title: Imitation Learning for Robots
   - local: bring_your_own_policies
-    title: Bring Your Own Policies
+    title: Adding a Policy
   - local: integrate_hardware
     title: Bring Your Own Hardware
   - local: hilserl
@@ -24,6 +26,12 @@
   - local: rename_map
     title: Using Rename Map and Empty Cameras
   title: "Tutorials"
+- sections:
+  - local: hardware_guide
+    title: Compute Hardware Guide
+  - local: torch_accelerators
+    title: PyTorch accelerators
+  title: "Compute & Hardware"
 - sections:
   - local: lerobot-dataset-v3
     title: Using LeRobotDataset
@@ -31,8 +39,14 @@
     title: Porting Large Datasets
   - local: using_dataset_tools
     title: Using the Dataset Tools
-  - local: dataset_subtask
-    title: Using Subtasks in the Dataset
+  - local: language_and_recipes
+    title: Language Columns and Recipes
+  - local: tools
+    title: Tools
+  - local: annotation_pipeline
+    title: Annotation Pipeline
+  - local: video_encoding_parameters
+    title: Video encoding parameters
   - local: streaming_video_encoding
     title: Streaming Video Encoding
   title: "Datasets"
@@ -47,6 +61,10 @@
     title: π₀-FAST (Pi0Fast)
   - local: pi05
     title: π₀.₅ (Pi05)
+  - local: molmoact2
+    title: MolmoAct2
+  - local: eo1
+    title: EO-1
   - local: groot
     title: NVIDIA GR00T N1.5
   - local: xvla
@@ -59,6 +77,10 @@
 - sections:
   - local: sarm
     title: SARM
+  - local: robometer
+    title: ROBOMETER
+  - local: topreward
+    title: TOPReward
   title: "Reward Models"
 - sections:
   - local: inference
@@ -131,6 +153,8 @@
     title: OMX
   - local: openarm
     title: OpenArm
+  - local: rebot_b601
+    title: reBot B601-DM
   title: "Robots"
 - sections:
   - local: phone_teleop
@@ -140,10 +164,6 @@
   - local: cameras
     title: Cameras
   title: "Sensors"
-- sections:
-  - local: torch_accelerators
-    title: PyTorch accelerators
-  title: "Supported Hardware"
 - sections:
   - local: notebooks
     title: Notebooks

docs/source/act.mdx

@@ -79,17 +79,13 @@ If your local computer doesn't have a powerful GPU, you can utilize Google Colab
 Once training is complete, you can evaluate your ACT policy using the `lerobot-record` command with your trained policy. This will run inference and record evaluation episodes:
 
 ```bash
-lerobot-record \
-  --robot.type=so100_follower \
+lerobot-rollout \
+  --strategy.type=base \
+  --policy.path=${HF_USER}/act_policy \
+  --robot.type=so101_follower \
   --robot.port=/dev/ttyACM0 \
-  --robot.id=my_robot \
   --robot.cameras="{ front: {type: opencv, index_or_path: 0, width: 640, height: 480, fps: 30}}" \
   --display_data=true \
-  --dataset.repo_id=${HF_USER}/eval_act_your_dataset \
-  --dataset.num_episodes=10 \
-  --dataset.single_task="Your task description" \
-  --dataset.streaming_encoding=true \
-  --dataset.encoder_threads=2 \
-  # --dataset.vcodec=auto \
-  --policy.path=${HF_USER}/act_policy
+  --task="Your task description" \ # can be skipped for ACT
+  --duration=60
 ```

docs/source/annotation_pipeline.mdx

@@ -0,0 +1,198 @@
+# Annotation Pipeline
+
+`lerobot-annotate` populates the two language columns introduced by the
+[Language Columns and Recipes](./language_and_recipes) page —
+`language_persistent` and `language_events` — directly into
+`data/chunk-*/file-*.parquet`.
+
+## What the pipeline produces
+
+A vocabulary-discovery phase derives a small canonical wording, then three
+modules write into a per-episode staging tree, then a single writer
+rewrites the data shards in place:
+
+| Style / atom                                | Column                | Module          |
+| ------------------------------------------- | --------------------- | --------------- |
+| `subtask` (Pi0.7-style "how, not what")     | `language_persistent` | `plan`          |
+| `plan` (initial + refresh on interjection)  | `language_persistent` | `plan`          |
+| `memory` (MEM-style compression)            | `language_persistent` | `plan`          |
+| `task_aug` (rephrasings of canonical task)  | `language_persistent` | `plan`          |
+| `interjection`                              | `language_events`     | `interjections` |
+| speech tool-call atom (`style=null`, `say`) | `language_events`     | `interjections` |
+| `vqa` (user / assistant pair)               | `language_events`     | `vqa`           |
+
+The `plan` module is constrained to a **canonical vocabulary** discovered
+once per dataset by the `vocabulary` module (phase 0). It watches a few
+sample episode videos (`--vocabulary.sample_episodes`, default `3`) and
+asks the VLM to derive a small set of imperative subtask labels and
+first-person memory milestones that recur across the demos. The VLM
+picks the right number of entries itself based on what it sees in the
+clips — short pick-and-place demos get ~6 subtask labels, longer
+multi-step recipes get more. The result lands at
+`meta/canonical_vocabulary.json` (human-readable / hand-editable) and
+is reused on every subsequent run. The `plan` module then constrains
+both subtask + memory generation to those exact strings — the
+downstream low-level policy sees a small, repeatable target
+distribution instead of thousands of LLM paraphrases. Disable with
+`--vocabulary.enabled=False` to fall back to free-form generation.
+
+The writer does **not** add a `tools` column to the parquet — the tool
+catalog lives at `meta/info.json["tools"]` instead (see
+[Tools](./tools)). After every annotation run the pipeline ensures the
+canonical `say` schema is present in that list, preserving any tools the
+user pre-declared.
+
+If you want to declare additional tools for a dataset before annotation
+runs, edit `meta/info.json["tools"]` directly — the pipeline preserves
+anything already there. Implementations of those tools live under
+`src/lerobot/tools/`; one file per tool, registered via
+`TOOL_REGISTRY`. See the [Tools](./tools) doc for the authoring guide.
+
+## Running locally
+
+Install the extra and invoke the console script. Episode-level
+concurrency comes from `--executor.episode_parallelism` (default 16);
+that is the only knob the in-process executor exposes.
+
+```bash
+uv sync --extra annotations
+uv run lerobot-annotate \
+  --root=/path/to/dataset \
+  --vlm.model_id=Qwen/Qwen2.5-VL-7B-Instruct
+```
+
+The pipeline attaches actual camera footage to every `plan` /
+`interjections` / `vqa` prompt by default, decoded from the dataset's
+first `observation.images.*` stream. Override with
+`--vlm.camera_key=observation.images.<name>` to pin a specific
+viewpoint. Datasets with no video tracks fall back to text-only prompts
+automatically.
+
+**The `plan` module sees the whole episode as one video block.** Subtask
+decomposition gets a `{"type":"video", "video":[<frames>]}` block
+covering the entire demonstration; Qwen-VL pools temporally on its own
+and decides where to cut. There is no keyframe stride or count knob —
+`--plan.max_video_frames` (default 128) only caps the frames packed
+into the video block as a model-capacity bound. The `interjections`
+module attaches a short window of frames straddling the interjection
+timestamp. The `vqa` module grounds each VQA pair on a single frame —
+its `--vqa.K` knob sets how many consecutive frames each emission tick
+anchors, and every anchored frame gets its own VQA pair on that one
+frame (there is no per-pair frame window).
+
+## Running on Hugging Face Jobs
+
+Distributed annotation is delegated to
+[Hugging Face Jobs](https://huggingface.co/docs/hub/en/jobs). The repo
+ships a launcher script you copy and edit for your dataset:
+
+```bash
+HF_TOKEN=hf_... uv run python examples/annotations/run_hf_job.py
+```
+
+[`examples/annotations/run_hf_job.py`](https://github.com/huggingface/lerobot/blob/main/examples/annotations/run_hf_job.py)
+spawns one `h200x2` job that:
+
+1. installs the branch under test plus the annotation extras,
+2. boots two vllm servers (one per GPU) for the chosen model,
+3. runs the `plan` / `interjections` / `vqa` modules across the dataset
+   via `lerobot-annotate`,
+4. uploads the annotated dataset to `--push_to_hub`.
+
+To target a different dataset, model, or hub repo, edit the `CMD` block
+inside the script — every flag in there maps directly onto a CLI flag of
+`lerobot-annotate` (see `lerobot-annotate --help` for the full list).
+
+## Style-to-recipe consumer mapping
+
+The pipeline's outputs are designed to be consumed by recipes (see
+[Language Columns and Recipes](./language_and_recipes)) — typically:
+
+- low-level / high-level / memory-update branches consume
+  `subtask`/`plan`/`memory` from `language_persistent`.
+- An interjection-response branch consumes `interjection` events plus
+  the paired speech atom (merged into one assistant target turn via
+  `tool_calls_from`) and the same-timestamp `plan` refresh.
+- A VQA branch consumes the `(vqa, user)` and `(vqa, assistant)` pairs
+  from `language_events`.
+
+## Why the design splits state from events
+
+Two things drive the scope:
+
+1. **Persistent state vs exact-event split.** Persistent rows
+   (`subtask`, `plan`, `memory`) broadcast per episode and answer "what
+   state is in force at this frame?". Event rows (`interjection`, `vqa`,
+   speech) only appear on the exact frame whose timestamp matches the
+   emission. The pipeline writes timestamps taken straight from the
+   source parquet — no floating-point recomputation.
+2. **One Qwen-VL pass.** All three modules share a single VLM client
+   (vLLM if available, transformers fallback) so the cost is one model
+   load per dataset, not three.
+
+## Module independence and staged reruns
+
+Each module writes its raw output to
+`<root>/.annotate_staging/episode_{N:06d}/<module>.jsonl`. That makes
+prompt iteration cheap — re-running one module overwrites only its own
+JSONL file before the writer composes the final parquet. Modules can be
+disabled via `--plan.enabled=false` (and likewise `--interjections.enabled`
+/ `--vqa.enabled`) to
+test them in isolation.
+
+## Validation/report checks before final write
+
+Before the writer runs, `StagingValidator` checks:
+
+- exact frame-timestamp alignment for every event row;
+- no orphan speech / interjection pairs;
+- `plan` is refreshed at every interjection timestamp;
+- `memory` rows fall on subtask boundaries (warning, not error);
+- VQA assistant `content` parses as JSON in one of the
+  bbox / keypoint / count / attribute / spatial shapes;
+- every row routes to the column dictated by `column_for_style(style)`.
+
+Errors abort the writer (`--skip_validation=true` overrides for debugging).
+
+## Paper inspirations per module
+
+- **`plan` module — subtasks.** Hi Robot ([Shi 2025](https://arxiv.org/abs/2502.19417))
+  atom granularity ("pick up one piece of lettuce", "place bowl to box");
+  Pi0.7 ([Physical Intelligence 2025](https://pi.website/pi07)) "how, not
+  what" detail.
+- **`plan` module — memory.** MEM ([Torne 2026](https://arxiv.org/abs/2603.03596))
+  compression directive: keep only minimal relevant information; functional
+  outcomes preserved, specific attributes dropped.
+- **`interjections` module.** Hi Robot scenario taxonomy: negative task,
+  situated correction, specific constraint, preference. Speech is a
+  tool-call-only atom (`tool_calls=[{type:function, function:{name:"say",
+arguments:{text:...}}}]`).
+- **`vqa` module.** ECoT ([Zawalski 2024](https://arxiv.org/abs/2407.08693))
+  grounded features (bounding boxes in pixel `[x_min, y_min, x_max, y_max]`,
+  keypoints) and Steerable VLA Policies ([Zhao 2025](https://arxiv.org/abs/2509.07626))
+  multi-abstraction grounding. Pi0.7 also grounds answers across
+  multiple abstraction levels.
+
+Future maintainers should adjust the prompt templates in
+`src/lerobot/annotations/steerable_pipeline/prompts/` against these
+references rather than rewriting from scratch.
+
+## Compute and list-size estimates
+
+Per episode, the pipeline issues O(`max_steps`) `plan`-module calls,
+O(`max_interjections_per_episode`) `interjections`-module calls, and
+O(`vqa_emission_hz × episode_seconds`) `vqa`-module calls. With defaults
+(8 subtasks, 1 interjection, 1 Hz × 3 pairs) and 30-second episodes, that
+is ~50 VLM calls per episode. `language_persistent` per episode is ~10s of
+KB at most (parquet dictionary-encodes one entry per episode);
+`language_events` is empty on most frames and is bounded by the number of
+emissions, not `num_frames × num_emissions`.
+
+## Reproducibility via seed and prompt hashes
+
+`--seed` (default 1729) feeds the per-episode RNGs that select interjection
+timestamps and VQA question types. Combined with the deterministic prompt
+templates checked into `prompts/`, two runs at the same seed against the
+same dataset and the same model checkpoint produce byte-identical staging
+artifacts. Prompt edits are recorded by file hash; future tooling can pin
+expected `(seed, prompt_hash)` pairs into the dataset card.

docs/source/bring_your_own_policies.mdx

@@ -1,60 +1,37 @@
-# Bring Your Own Policies
+# Adding a Policy
 
-This tutorial explains how to integrate your own custom policy implementations into the LeRobot ecosystem, allowing you to leverage all LeRobot tools for training, evaluation, and deployment while using your own algorithms.
+This guide walks you through implementing a custom policy and getting it to work with LeRobot's training, evaluation, and deployment tools. There are two paths:
 
-## Step 1: Create a Policy Package
+- **Plugin (out-of-tree)** — ship your policy as a standalone `lerobot_policy_*` package. Faster, no PR required, easy to iterate. Right for experimentation, internal use, or when you want to publish independently.
+- **In-tree (contributed to LeRobot)** — land your policy directly in `src/lerobot/policies/`. Requires a PR, but makes your policy a first-class citizen of the library.
 
-Your custom policy should be organized as an installable Python package following LeRobot's plugin conventions.
+The plugin route is usually the right starting point — promote to in-tree once the policy has stabilized and there's clear value in shipping it with the library.
 
-### Package Structure
+Either way, the building blocks are the same: a configuration class, a policy class, and a processor factory. The first half of this guide covers those shared pieces; the second half covers the path-specific scaffolding ([Path A](#path-a-out-of-tree-plugin), [Path B](#path-b-contributing-in-tree)).
 
-Create a package with the prefix `lerobot_policy_` (IMPORTANT!) followed by your policy name:
+A note on tone: robot-learning is an actively evolving field, and "what a policy looks like" can shift with each new architecture. The conventions described here exist because they let `lerobot-train` and `lerobot-eval` work uniformly across very different models. When a new policy genuinely doesn't fit them, raise it (in your PR, or an issue) — the conventions are not sacred.
 
-```bash
-lerobot_policy_my_custom_policy/
-├── pyproject.toml
-└── src/
-    └── lerobot_policy_my_custom_policy/
-        ├── __init__.py
-        ├── configuration_my_custom_policy.py
-        ├── modeling_my_custom_policy.py
-        └── processor_my_custom_policy.py
-```
+---
 
-### Package Configuration
+## Anatomy of a policy
 
-Set up your `pyproject.toml`:
+Three building blocks make up every policy. The names below use `my_policy` as a placeholder — replace with your policy's name. That name is load-bearing: it must match the string you pass to `@PreTrainedConfig.register_subclass`, the `MyPolicy.name` class attribute, and the `make_<name>_pre_post_processors` factory function (more on each below).
 
-```toml
-[project]
-name = "lerobot_policy_my_custom_policy"
-version = "0.1.0"
-dependencies = [
-    # your policy-specific dependencies
-]
-requires-python = ">= 3.12"
+### Configuration class
 
-[build-system]
-build-backend = # your-build-backend
-requires = # your-build-system
-```
-
-## Step 2: Define the Policy Configuration
-
-Create a configuration class that inherits from [`PreTrainedConfig`](https://github.com/huggingface/lerobot/blob/main/src/lerobot/configs/policies.py) and registers your policy type:
-Here is a template to get you started, customize the parameters and methods as needed for your policy's architecture and training requirements.
+Inherit from [`PreTrainedConfig`](https://github.com/huggingface/lerobot/blob/main/src/lerobot/configs/policies.py) and register your policy type. Here is a template — customize the parameters and methods as needed for your policy's architecture and training requirements.
 
 ```python
-# configuration_my_custom_policy.py
+# configuration_my_policy.py
 from dataclasses import dataclass, field
 from lerobot.configs import PreTrainedConfig
 from lerobot.optim import AdamWConfig
 from lerobot.optim import CosineDecayWithWarmupSchedulerConfig
 
-@PreTrainedConfig.register_subclass("my_custom_policy")
+@PreTrainedConfig.register_subclass("my_policy")
 @dataclass
-class MyCustomPolicyConfig(PreTrainedConfig):
-    """Configuration class for MyCustomPolicy.
+class MyPolicyConfig(PreTrainedConfig):
+    """Configuration class for MyPolicy.
 
     Args:
         n_obs_steps: Number of observation steps to use as input
@@ -77,16 +54,20 @@ class MyCustomPolicyConfig(PreTrainedConfig):
             raise ValueError("n_action_steps cannot exceed horizon")
 
     def validate_features(self) -> None:
-        """Validate input/output feature compatibility."""
+        """Validate input/output feature compatibility.
+
+        Call this explicitly from your policy's __init__ — the base class does not.
+        """
         if not self.image_features:
-            raise ValueError("MyCustomPolicy requires at least one image feature.")
+            raise ValueError("MyPolicy requires at least one image feature.")
         if self.action_feature is None:
-            raise ValueError("MyCustomPolicy requires 'action' in output_features.")
+            raise ValueError("MyPolicy requires 'action' in output_features.")
 
     def get_optimizer_preset(self) -> AdamWConfig:
         return AdamWConfig(lr=self.optimizer_lr, weight_decay=self.optimizer_weight_decay)
 
     def get_scheduler_preset(self):
+        """Return a LRSchedulerConfig from lerobot.optim, or None."""
         return None
 
     @property
@@ -101,8 +82,7 @@ class MyCustomPolicyConfig(PreTrainedConfig):
 
     @property
     def action_delta_indices(self) -> list[int]:
-        """Relative timestep offsets for the action chunk the dataset loader returns.
-        """
+        """Relative timestep offsets for the action chunk the dataset loader returns."""
         return list(range(self.horizon))
 
     @property
@@ -110,32 +90,34 @@ class MyCustomPolicyConfig(PreTrainedConfig):
         return None
 ```
 
-## Step 3: Implement the Policy Class
+The string you pass to `@register_subclass` must match `MyPolicy.name` (next section) and is what users supply as `--policy.type` on the CLI. Default to `AdamW` from `lerobot.optim` for `get_optimizer_preset` unless you genuinely need otherwise.
 
-Create your policy implementation by inheriting from [`PreTrainedPolicy`](https://github.com/huggingface/lerobot/blob/main/src/lerobot/policies/pretrained.py):
+### Policy class
+
+Inherit from [`PreTrainedPolicy`](https://github.com/huggingface/lerobot/blob/main/src/lerobot/policies/pretrained.py) and set two class attributes — both are checked by `__init_subclass__`:
 
 ```python
-# modeling_my_custom_policy.py
+# modeling_my_policy.py
 import torch
 import torch.nn as nn
 from typing import Any
 
 from lerobot.policies import PreTrainedPolicy
 from lerobot.utils.constants import ACTION
-from .configuration_my_custom_policy import MyCustomPolicyConfig
+from .configuration_my_policy import MyPolicyConfig
 
-class MyCustomPolicy(PreTrainedPolicy):
-    config_class = MyCustomPolicyConfig  # must match the string in @register_subclass
-    name = "my_custom_policy"
+class MyPolicy(PreTrainedPolicy):
+    config_class = MyPolicyConfig  # must match the string in @register_subclass
+    name = "my_policy"
 
-    def __init__(self, config: MyCustomPolicyConfig, dataset_stats: dict[str, Any] = None):
+    def __init__(self, config: MyPolicyConfig, dataset_stats: dict[str, Any] = None):
         super().__init__(config, dataset_stats)
         config.validate_features()  # not called automatically by the base class
         self.config = config
         self.model = ...  # your nn.Module here
 
     def reset(self):
-        """Reset episode state."""
+        """Reset per-episode state. Called by lerobot-eval at the start of each episode."""
         ...
 
     def get_optim_params(self) -> dict:
@@ -147,35 +129,51 @@ class MyCustomPolicy(PreTrainedPolicy):
         ...
 
     def select_action(self, batch: dict[str, torch.Tensor], **kwargs) -> torch.Tensor:
-        """Return a single action for the current timestep (called at inference)."""
+        """Return a single action for the current timestep (called every step at inference)."""
         ...
 
-    def forward(self, batch: dict[str, torch.Tensor]) -> dict[str, torch.Tensor]:
+    def forward(self, batch: dict[str, torch.Tensor]) -> tuple[torch.Tensor, dict | None]:
         """Compute the training loss.
 
+        Returns `(loss, output_dict)`. `output_dict` may be `None`; everything in it must be
+        logging-friendly Python natives (no tensors with gradients).
+
         `batch["action_is_pad"]` is a bool mask of shape (B, horizon) that marks
-        timesteps padded because the episode ended before `horizon` steps, you
+        timesteps padded because the episode ended before `horizon` steps; you
         can exclude those from your loss.
         """
         actions = batch[ACTION]
         action_is_pad = batch.get("action_is_pad")
         ...
-        return {"loss": ...}
+        return loss, {"some_loss_component": some_loss_component.item()}
 ```
 
-## Step 4: Add Data Processors
+The methods called by the train/eval loops:
 
-Create processor functions. For a concrete reference, see [processor_act.py](https://github.com/huggingface/lerobot/blob/main/src/lerobot/policies/act/processor_act.py) or [processor_diffusion.py](https://github.com/huggingface/lerobot/blob/main/src/lerobot/policies/diffusion/processor_diffusion.py).
+| Method                                                            | Used by           | What it does                                                                                                                                                                                                                                         |
+| ----------------------------------------------------------------- | ----------------- | ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
+| `reset() -> None`                                                 | `lerobot-eval`    | Clear per-episode state at the start of each episode.                                                                                                                                                                                                |
+| `select_action(batch, **kwargs) -> Tensor`                        | `lerobot-eval`    | Return the next action `(B, action_dim)`. Called every step.                                                                                                                                                                                         |
+| `predict_action_chunk(batch, **kwargs) -> Tensor`                 | the policy itself | Return an action chunk `(B, chunk_size, action_dim)`. Currently abstract on the base class — raise `NotImplementedError` if your policy doesn't chunk.                                                                                               |
+| `forward(batch, reduction="mean") -> tuple[Tensor, dict \| None]` | `lerobot-train`   | Return `(loss, output_dict)`. Accept `reduction="none"` if you want to support per-sample weighting.                                                                                                                                                 |
+| `get_optim_params() -> dict`                                      | the optimizer     | Return `self.parameters()` for simple policies; return a named parameter dict for [multi-optimizer policies](https://github.com/huggingface/lerobot/blob/ecd38c50d7d15b4184cf42649ff1185ee2e11eeb/src/lerobot/policies/sac/modeling_sac.py#L61-L73). |
+| `update() -> None` _(optional)_                                   | `lerobot-train`   | Called after each optimizer step _if defined_. Use for EMA, target nets, replay buffers (TDMPC uses this).                                                                                                                                           |
+
+Batches are flat dictionaries keyed by the constants in [`lerobot.utils.constants`](https://github.com/huggingface/lerobot/blob/main/src/lerobot/utils/constants.py): `OBS_STATE` (`observation.state.<motor>`), `OBS_IMAGES` (`observation.images.<camera>`), `OBS_LANGUAGE`, `ACTION`, etc. Reuse the constants — don't invent new prefixes.
+
+### Processor functions
+
+LeRobot uses `PolicyProcessorPipeline`s to normalize inputs and de-normalize outputs around your policy. For a concrete reference, see [`processor_act.py`](https://github.com/huggingface/lerobot/blob/main/src/lerobot/policies/act/processor_act.py) or [`processor_diffusion.py`](https://github.com/huggingface/lerobot/blob/main/src/lerobot/policies/diffusion/processor_diffusion.py).
 
 ```python
-# processor_my_custom_policy.py
+# processor_my_policy.py
 from typing import Any
 import torch
 
 from lerobot.processor import PolicyAction, PolicyProcessorPipeline
 
 
-def make_my_custom_policy_pre_post_processors(
+def make_my_policy_pre_post_processors(
     config,
     dataset_stats: dict[str, dict[str, torch.Tensor]] | None = None,
 ) -> tuple[
@@ -187,11 +185,48 @@ def make_my_custom_policy_pre_post_processors(
     return preprocessor, postprocessor
 ```
 
-**Important - function naming:** LeRobot discovers your processor by name. The function **must** be called `make_{policy_name}_pre_post_processors` (matching the string you passed to `@PreTrainedConfig.register_subclass`).
+**Important — function naming:** LeRobot discovers your processor by name. The function **must** be called `make_{policy_name}_pre_post_processors` (matching the string you passed to `@PreTrainedConfig.register_subclass`).
 
-## Step 5: Package Initialization
+---
 
-Expose your classes in the package's `__init__.py`:
+## Path A: Out-of-tree plugin
+
+The fastest way to ship a policy: package it as a standalone Python distribution and install it alongside LeRobot. No PR required, you own the release cycle, and you can publish to PyPI under your own namespace.
+
+### Package structure
+
+Create a package with the prefix `lerobot_policy_` (IMPORTANT!) followed by your policy name:
+
+```bash
+lerobot_policy_my_policy/
+├── pyproject.toml
+└── src/
+    └── lerobot_policy_my_policy/
+        ├── __init__.py
+        ├── configuration_my_policy.py
+        ├── modeling_my_policy.py
+        └── processor_my_policy.py
+```
+
+### `pyproject.toml`
+
+```toml
+[project]
+name = "lerobot_policy_my_policy"
+version = "0.1.0"
+dependencies = [
+    # your policy-specific dependencies
+]
+requires-python = ">= 3.12"
+
+[build-system]
+build-backend = # your-build-backend
+requires = # your-build-system
+```
+
+### Package `__init__.py`
+
+Expose your classes in the package's `__init__.py` and guard against missing `lerobot`:
 
 ```python
 # __init__.py
@@ -204,44 +239,148 @@ except ImportError:
         "lerobot is not installed. Please install lerobot to use this policy package."
     )
 
-from .configuration_my_custom_policy import MyCustomPolicyConfig
-from .modeling_my_custom_policy import MyCustomPolicy
-from .processor_my_custom_policy import make_my_custom_policy_pre_post_processors
+from .configuration_my_policy import MyPolicyConfig
+from .modeling_my_policy import MyPolicy
+from .processor_my_policy import make_my_policy_pre_post_processors
 
 __all__ = [
-    "MyCustomPolicyConfig",
-    "MyCustomPolicy",
-    "make_my_custom_policy_pre_post_processors",
+    "MyPolicyConfig",
+    "MyPolicy",
+    "make_my_policy_pre_post_processors",
 ]
 ```
 
-## Step 6: Installation and Usage
-
-### Install Your Policy Package
+### Install and use
 
 ```bash
-cd lerobot_policy_my_custom_policy
+cd lerobot_policy_my_policy
 pip install -e .
 
 # Or install from PyPI if published
-pip install lerobot_policy_my_custom_policy
+pip install lerobot_policy_my_policy
 ```
 
-### Use Your Policy
-
 Once installed, your policy automatically integrates with LeRobot's training and evaluation tools:
 
 ```bash
 lerobot-train \
-    --policy.type my_custom_policy \
+    --policy.type my_policy \
     --env.type pusht \
     --steps 200000
 ```
 
-## Examples and Community Contributions
+---
+
+## Path B: Contributing in-tree
+
+When your policy has stabilized and there's clear value in shipping it with the library, you can land it directly in LeRobot. Read the general [contribution guide](./contributing) and the [PR template](https://github.com/huggingface/lerobot/blob/main/.github/PULL_REQUEST_TEMPLATE.md) first — that's where you'll find the testing/quality expectations every PR has to meet (`pre-commit run -a`, `pytest`, the community-review rule, etc.). What's below is the policy-specific layer on top of that.
+
+### In-tree layout
+
+```
+src/lerobot/policies/my_policy/
+├── __init__.py                    # re-exports config + modeling + processor factory
+├── configuration_my_policy.py     # MyPolicyConfig + @register_subclass
+├── modeling_my_policy.py          # MyPolicy(PreTrainedPolicy)
+├── processor_my_policy.py         # make_my_policy_pre_post_processors
+└── README.md                      # symlink → ../../../../docs/source/policy_my_policy_README.md
+```
+
+Two notes:
+
+- The `README.md` next to the source is a **symlink** into `docs/source/policy_<name>_README.md` — the actual file lives under `docs/`. Existing policies (act, smolvla, diffusion, …) all do this; copy one of those symlinks. The policy README is conventionally minimal: paper link + BibTeX citation.
+- The user-facing tutorial — what to install, how to train, hyperparameters, benchmark numbers — lives separately at `docs/source/<my_policy>.mdx` and is registered in `_toctree.yml` under "Policies".
+
+The file names are load-bearing: the factory does lazy imports by name, and the processor is discovered by the `make_<policy_name>_pre_post_processors` convention.
+
+### Wiring
+
+Three places need to know about your policy. All by name.
+
+1. **`policies/__init__.py`** — re-export `MyPolicyConfig` and add it to `__all__`. **Don't** re-export the modeling class; it loads lazily through the factory (so `import lerobot` stays fast).
+2. **`factory.py:get_policy_class`** — add a branch returning `MyPolicy` from a lazy import.
+3. **`factory.py:make_policy_config`** and **`factory.py:make_pre_post_processors`** — same idea, two more branches.
+
+Mirror an existing policy that's structurally similar to yours; the diff is small.
+
+### Heavy / optional dependencies
+
+Most policies need a heavy backbone (transformers, diffusers, a specific VLM SDK). The convention is **two-step gating**: a `TYPE_CHECKING`-guarded import at module top, and a `require_package` runtime check in the constructor. [`modeling_diffusion.py`](https://github.com/huggingface/lerobot/blob/main/src/lerobot/policies/diffusion/modeling_diffusion.py) is the canonical reference:
+
+```python
+from typing import TYPE_CHECKING
+from lerobot.utils.import_utils import _diffusers_available, require_package
+
+if TYPE_CHECKING or _diffusers_available:
+    from diffusers.schedulers.scheduling_ddim import DDIMScheduler
+else:
+    DDIMScheduler = None  # keeps the symbol bindable at import time
+
+class DiffusionPolicy(PreTrainedPolicy):
+    def __init__(self, config):
+        require_package("diffusers", extra="diffusion")
+        super().__init__(config)
+        ...
+```
+
+This way:
+
+- `import lerobot.policies` keeps working without the extra installed (the symbol is just bound to `None`).
+- Type checkers see the real symbol.
+- Instantiating the policy without the extra raises a clear `ImportError` pointing at `pip install 'lerobot[diffusion]'`.
+
+Add a matching extra to [`pyproject.toml`](https://github.com/huggingface/lerobot/blob/main/pyproject.toml) `[project.optional-dependencies]` and include it in the `all` extra so `pip install 'lerobot[all]'` keeps installing everything.
+
+### Benchmarks and a published checkpoint
+
+A new policy is much easier to review — and far more useful — when it ships with a working checkpoint and at least one number you can reproduce.
+
+**Pick at least one in-tree benchmark.** LeRobot ships sim benchmarks with per-benchmark Docker images (LIBERO, LIBERO-plus, Meta-World, RoboTwin 2.0, RoboCasa365, RoboCerebra, RoboMME, VLABench and more). Pick the one that matches your policy's modality — VLAs usually go to LIBERO or VLABench; image-only BC to LIBERO or Meta-World. The full list lives under [Benchmarks](./libero) in the docs sidebar.
+
+**Push the checkpoint & processors** to the Hub under `lerobot/<policy>_<benchmark>` (or your namespace if you don't have write access; a maintainer can mirror it). Use `PreTrainedPolicy.push_model_to_hub` so the repo gets `config.json`, `model.safetensors`, and a model card.
+
+**Report results in your policy's MDX**, with the exact `lerobot-eval` command and hardware so anyone can re-run:
+
+```markdown
+## Results
+
+Evaluated on LIBERO with `lerobot/<policy>_libero`:
+
+| Suite          | Success rate | n_episodes |
+| -------------- | -----------: | ---------: |
+| libero_spatial |        87.5% |         50 |
+| libero_object  |        93.0% |         50 |
+| libero_goal    |        81.5% |         50 |
+| libero_10      |        62.0% |         50 |
+| **average**    |    **81.0%** |        200 |
+
+Reproduce: `lerobot-eval --policy.path=lerobot/<policy>_libero --env.type=libero --env.task=libero_spatial --eval.n_episodes=50` (1× A100 40 GB).
+```
+
+Use `n_episodes ≥ 50` per suite for stable success-rate estimates.
+
+If your policy is real-robot-only and no sim benchmark applies, swap the sim eval for: a public training dataset on the Hub, the `lerobot-train` command, the checkpoint, and a real-robot success rate over ≥10 episodes via `lerobot-rollout --policy.path=...`.
+
+### PR checklist
+
+The general expectations are in [`CONTRIBUTING.md`](https://github.com/huggingface/lerobot/blob/main/CONTRIBUTING.md) and the [PR template](https://github.com/huggingface/lerobot/blob/main/.github/PULL_REQUEST_TEMPLATE.md). On top of those, reviewers will look for:
+
+- [ ] `MyPolicy` and `MyPolicyConfig` cover the surface above; `__init_subclass__` accepts the class.
+- [ ] `factory.py` and `policies/__init__.py` are wired (lazy imports for modeling).
+- [ ] `make_my_policy_pre_post_processors` follows the naming convention.
+- [ ] Optional deps live behind a `[project.optional-dependencies]` extra and the `TYPE_CHECKING + require_package` guard.
+- [ ] `tests/policies/` updated; backward-compat artifact committed & policy-specific tests.
+- [ ] `src/lerobot/policies/<name>/README.md` symlinked into `docs/source/policy_<name>_README.md`; user-facing `docs/source/<name>.mdx` written and added to `_toctree.yml`.
+- [ ] At least one reproducible benchmark eval in the policy MDX with a published checkpoint (sim benchmark, or real-robot dataset + checkpoint).
+
+The fastest way to get a clean PR is to copy the directory of the existing policy closest to yours, rename, and replace contents method by method. Don't wait until everything is polished — open a draft PR early and iterate with us; reviewers would much rather give feedback on a half-finished branch than a fully-merged one.
+
+---
+
+## Examples and community contributions
 
 Check out these example policy implementations:
 
-- [DiTFlow Policy](https://github.com/danielsanjosepro/lerobot_policy_ditflow) - Diffusion Transformer policy with flow-matching objective. Try it out in this example: [DiTFlow Example](https://github.com/danielsanjosepro/test_lerobot_policy_ditflow)
+- [DiTFlow Policy](https://github.com/danielsanjosepro/lerobot_policy_ditflow) — Diffusion Transformer policy with flow-matching objective. Try it out in this example: [DiTFlow Example](https://github.com/danielsanjosepro/test_lerobot_policy_ditflow)
 
-Share your policy implementations with the community! 🤗
+Thanks for taking the time to bring a new policy into LeRobot. Every architecture that lands in `main` — and every plugin published by the community — makes the library a little more useful for the next person, and a little more representative of where robot learning is going. We're looking forward to seeing what you ship. 🤗

docs/source/cheat-sheet.mdx

@@ -0,0 +1,139 @@
+# Cheat sheet
+
+All of the LeRobot commands in one place. If you forgot how to use a specific command or want to learn about a new one you can do it here.
+
+> [!WARNING]
+> For all of the commands listed below remember to change the ports/names/ids to your own values!
+
+> [!TIP]
+> Another great way to look at all the commands and get them configured for your specific setup is to use this [Jupyter Notebook](https://github.com/huggingface/lerobot/blob/main/examples/notebooks/quickstart.ipynb).
+
+### Setup and installation
+
+For installation please look at [LeRobot Installation](https://huggingface.co/docs/lerobot/main/en/installation).
+
+### Useful tools
+
+###### Find port
+
+Use this to identify which serial ports your robots are connected to. Follow the instructions in your terminal: you will be asked to unplug the USB cable and press Enter. The script will then detect and print the correct serial port for that robot.
+
+```bash
+lerobot-find-port
+```
+
+###### Find cameras
+
+Quickly find camera indices and verify their output. This command prints camera information to the terminal and saves test frames from each detected camera to `lerobot/outputs/captured_images`
+
+```bash
+lerobot-find-cameras
+```
+
+### Calibration
+
+In most cases you will need to perform calibration just once for each robot and teleoperation device. Before performing the calibration make sure that all the joints are roughly in the middle position.
+
+```bash
+lerobot-calibrate \
+    --robot.type=so101_follower \
+    --robot.port=/dev/ttyACM0 \
+    --robot.id=my_follower_arm
+```
+
+Make sure that you use the same IDs used during calibration later for the other scripts. That's how LeRobot finds the calibration files.
+
+### Teleoperation
+
+Teleoperating with two cameras and displaying the data with Rerun.
+
+```bash
+lerobot-teleoperate \
+    --robot.type=so101_follower \
+    --robot.port=/dev/ttyACM0 \
+    --robot.id=my_follower_arm \
+    --robot.cameras="{ top: {type: opencv, index_or_path: 1, width: 640, height: 480, fps: 30}, wrist: {type: opencv, index_or_path: 0, width: 640, height: 480, fps: 30} }" \
+    --teleop.type=so101_leader \
+    --teleop.port=/dev/ttyACM1 \
+    --teleop.id=my_leader_arm \
+    --display_data=true
+```
+
+### Recording a dataset
+
+The dataset is automatically uploaded to the server and saved under repo_id, make sure you are logged in to your HF account with CLI:
+`hf auth login`
+
+You can get the token from: [https://huggingface.co/settings/tokens](https://huggingface.co/settings/tokens)
+
+```bash
+lerobot-record \
+    --robot.type=so101_follower \
+    --robot.port=/dev/ttyACM0 \
+    --robot.id=my_follower_arm \
+    --robot.cameras="{ top: {type: opencv, index_or_path: 1, width: 640, height: 480, fps: 30}, wrist: {type: opencv, index_or_path: 0, width: 640, height: 480, fps: 30} }" \
+    --teleop.type=so101_leader \
+    --teleop.port=/dev/ttyACM1 \
+    --teleop.id=my_leader_arm \
+    --dataset.repo_id=${HF_USER}/so101_dataset_test \
+    --dataset.num_episodes=30 \
+    --dataset.single_task="put the red brick in a bowl" \
+    --dataset.streaming_encoding=true \
+    --display_data=true
+```
+
+While collecting the dataset you can control the process with your keyboard:
+Control the data recording flow using keyboard shortcuts:
+
+- Press **Right Arrow (`→`)**: Save episode and move to the next.
+- Press **Left Arrow (`←`)**: Delete current episode and retry.
+- Press **Escape (`ESC`)**: Stop, encode videos, and upload.
+
+### Training
+
+Depending on your hardware training the policy might take a few hours. That's how you train simple `ACT` policy:
+
+```bash
+lerobot-train \
+    --dataset.repo_id=${HF_USER}/so101_dataset_test \
+    --policy.type=act \
+    --output_dir=outputs/train/act_so101_test \
+    --job_name=act_so101_test \
+    --policy.device=cuda \
+    --wandb.enable=true \
+    --policy.repo_id=${HF_USER}/policy_test \
+    --steps=20000
+```
+
+- Policy Types: `act`, `diffusion`, `smolvla`, `pi05`
+- Devices: `cuda` (NVIDIA), `mps` (Apple Silicon), `cpu`
+
+If you want to fine-tune a specific model you can provide the path to the model. In this case path is enough and type can be skipped.
+
+```bash
+lerobot-train \
+    --dataset.repo_id=${HF_USER}/so101_dataset_test \
+    --policy.path=username/the_policy_to_finetune \
+    --policy.device=cuda \
+    --policy.repo_id=${HF_USER}/policy_test \
+    --output_dir=outputs/train/act_so101_test \
+    --steps=20000
+```
+
+### Inference
+
+Inference means running the trained policy/model on a robot. For that we use `lerobot-rollout`. You will need to provide a path to your policy. It can be a local path or a path to Hugging Face for example "lerobot/folding_latest". Your cameras configuration needs to match what was used when collecting the dataset. Duration is in seconds if unspecified, it will run forever.
+
+> [!TIP]
+> If you are using the previous release V0.5.1 instead of `lerobot-rollout` you need to use `lerobot-record`. More information [here](https://huggingface.co/docs/lerobot/v0.5.1/en/il_robots#run-inference-and-evaluate-your-policy).
+
+```bash
+lerobot-rollout \
+    --strategy.type=base \
+    --policy.path=${HF_USER}/my_policy \
+    --robot.type=so101_follower \
+    --robot.port=/dev/ttyACM1 \
+    --robot.cameras="{ up: {type: opencv, index_or_path: /dev/video1, width: 640, height: 480, fps: 30}, side: {type: opencv, index_or_path: /dev/video5, width: 640, height: 480, fps: 30}}" \
+    --task="Put lego brick into the transparent box" \
+    --duration=60
+```

docs/source/dataset_subtask.mdx

@@ -1,277 +0,0 @@
-# Using Subtasks in LeRobot Datasets
-
-Subtask support in robotics datasets has proven effective in improving robot reasoning and understanding. Subtasks are particularly useful for:
-
-- **Hierarchical policies**: Building policies that include subtask predictions to visualize robot reasoning in real time
-- **Reward modeling**: Helping reward models understand task progression (e.g., SARM-style stage-aware reward models)
-- **Task decomposition**: Breaking down complex manipulation tasks into atomic, interpretable steps
-
-LeRobotDataset now supports subtasks as part of its dataset structure, alongside tasks.
-
-## What are Subtasks?
-
-While a **task** describes the overall goal (e.g., "Pick up the apple and place it in the basket"), **subtasks** break down the execution into finer-grained steps:
-
-1. "Approach the apple"
-2. "Grasp the apple"
-3. "Lift the apple"
-4. "Move to basket"
-5. "Release the apple"
-
-Each frame in the dataset can be annotated with its corresponding subtask, enabling models to learn and predict these intermediate stages.
-
-<img
-  src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/lerobot/subtask-asset.png"
-  alt="An overview of subtask annotation showing how frames are labeled with intermediate subtask stages"
-  width="80%"
-/>
-
-<p>
-  <em>Figure: Overview of subtask annotation.</em>
-</p>
-
-**Reference:** _Subtask-learning based for robot self-assembly in flexible collaborative assembly in manufacturing_, Original Article, Published: 19 April 2022.
-
-## Dataset Structure
-
-Subtask information is stored in the dataset metadata:
-
-```
-my-dataset/
-├── data/
-│   └── ...
-├── meta/
-│   ├── info.json
-│   ├── stats.json
-│   ├── tasks.parquet
-│   ├── subtasks.parquet      # Subtask index → subtask string mapping
-│   └── episodes/
-│       └── ...
-└── videos/
-    └── ...
-```
-
-### Subtasks Parquet File
-
-The `meta/subtasks.parquet` file maps subtask indices to their natural language descriptions:
-
-| subtask_index | subtask (index column) |
-| ------------- | ---------------------- |
-| 0             | "Approach the apple"   |
-| 1             | "Grasp the apple"      |
-| 2             | "Lift the apple"       |
-| ...           | ...                    |
-
-### Frame-Level Annotations
-
-Each frame in the dataset can include a `subtask_index` field that references the subtasks parquet file:
-
-```python
-# Example frame data in the parquet file
-{
-    "index": 42,
-    "timestamp": 1.4,
-    "episode_index": 0,
-    "task_index": 0,
-    "subtask_index": 2,  # References "Lift the apple"
-    "observation.state": [...],
-    "action": [...],
-}
-```
-
-## Annotating Datasets with Subtasks
-
-We provide a HuggingFace Space for easily annotating any LeRobotDataset with subtasks:
-
-**[https://huggingface.co/spaces/lerobot/annotate](https://huggingface.co/spaces/lerobot/annotate)**
-
-After completing your annotation:
-
-1. Click "Push to Hub" to upload your annotated dataset
-2. You can also run the annotation space locally by following the instructions at [github.com/huggingface/lerobot-annotate](https://github.com/huggingface/lerobot-annotate)
-
-## Loading Datasets with Subtasks
-
-When you load a dataset with subtask annotations, the subtask information is automatically available:
-
-```python
-from lerobot.datasets import LeRobotDataset
-
-# Load a dataset with subtask annotations
-dataset = LeRobotDataset("jadechoghari/collect-fruit-annotated")
-
-# Access a sample
-sample = dataset[100]
-
-# The sample includes both task and subtask information
-print(sample["task"])        # "Collect the fruit"
-print(sample["subtask"])     # "Grasp the apple"
-print(sample["task_index"])  # tensor(0)
-print(sample["subtask_index"])  # tensor(2)
-```
-
-### Checking for Subtask Support
-
-You can check if a dataset has subtask annotations:
-
-```python
-# Check if subtasks are available
-has_subtasks = (
-    "subtask_index" in dataset.features
-    and dataset.meta.subtasks is not None
-)
-
-if has_subtasks:
-    print(f"Dataset has {len(dataset.meta.subtasks)} unique subtasks")
-    print("Subtasks:", list(dataset.meta.subtasks.index))
-```
-
-## Using Subtasks for Training
-
-### With the Tokenizer Processor
-
-The `TokenizerProcessor` automatically handles subtask tokenization for Vision-Language Action (VLA) models:
-
-```python
-from lerobot.processor import TokenizerProcessorStep
-
-# Create a tokenizer processor step
-tokenizer_processor = TokenizerProcessorStep(
-    tokenizer_name_or_path="google/paligemma-3b-pt-224",
-    padding="max_length",
-    max_length=64,
-)
-
-# The processor will automatically tokenize subtasks if present in the batch
-# and add them to the observation under:
-# - "observation.subtask.tokens"
-# - "observation.subtask.attention_mask"
-```
-
-When subtasks are available in the batch, the tokenizer processor adds:
-
-- `observation.subtask.tokens`: Tokenized subtask text
-- `observation.subtask.attention_mask`: Attention mask for the subtask tokens
-
-### DataLoader with Subtasks
-
-```python
-import torch
-from lerobot.datasets import LeRobotDataset
-
-dataset = LeRobotDataset("jadechoghari/collect-fruit-annotated")
-
-dataloader = torch.utils.data.DataLoader(
-    dataset,
-    batch_size=16,
-    shuffle=True,
-)
-
-for batch in dataloader:
-    # Access subtask information in the batch
-    subtasks = batch["subtask"]  # List of subtask strings
-    subtask_indices = batch["subtask_index"]  # Tensor of subtask indices
-
-    # Use for training hierarchical policies or reward models
-    print(f"Batch subtasks: {set(subtasks)}")
-```
-
-## Example Datasets with Subtask Annotations
-
-Try loading a dataset with subtask annotations:
-
-```python
-from lerobot.datasets import LeRobotDataset
-
-# Example dataset with subtask annotations
-dataset = LeRobotDataset("jadechoghari/collect-fruit-annotated")
-
-# Explore the subtasks
-print("Available subtasks:")
-for subtask_name in dataset.meta.subtasks.index:
-    print(f"  - {subtask_name}")
-
-# Get subtask distribution
-subtask_counts = {}
-for i in range(len(dataset)):
-    sample = dataset[i]
-    subtask = sample["subtask"]
-    subtask_counts[subtask] = subtask_counts.get(subtask, 0) + 1
-
-print("\nSubtask distribution:")
-for subtask, count in sorted(subtask_counts.items(), key=lambda x: -x[1]):
-    print(f"  {subtask}: {count} frames")
-```
-
-## Use Cases
-
-### 1. Hierarchical Policy Training
-
-Train policies that predict both actions and current subtask:
-
-```python
-class HierarchicalPolicy(nn.Module):
-    def __init__(self, num_subtasks):
-        super().__init__()
-        self.action_head = nn.Linear(hidden_dim, action_dim)
-        self.subtask_head = nn.Linear(hidden_dim, num_subtasks)
-
-    def forward(self, observations):
-        features = self.encoder(observations)
-        actions = self.action_head(features)
-        subtask_logits = self.subtask_head(features)
-        return actions, subtask_logits
-```
-
-### 2. Stage-Aware Reward Modeling (SARM)
-
-Build reward models that understand task progression:
-
-```python
-# SARM predicts:
-# - Stage: Which subtask is being executed (discrete)
-# - Progress: How far along the subtask (continuous 0-1)
-
-class SARMRewardModel(nn.Module):
-    def forward(self, observations):
-        features = self.encoder(observations)
-        stage_logits = self.stage_classifier(features)
-        progress = self.progress_regressor(features)
-        return stage_logits, progress
-```
-
-### 3. Progress Visualization
-
-Monitor robot execution by tracking subtask progression:
-
-```python
-def visualize_execution(model, observations):
-    for t, obs in enumerate(observations):
-        action, subtask_logits = model(obs)
-        predicted_subtask = subtask_names[subtask_logits.argmax()]
-        print(f"t={t}: Executing '{predicted_subtask}'")
-```
-
-## API Reference
-
-### LeRobotDataset Properties
-
-| Property                    | Type                   | Description                                |
-| --------------------------- | ---------------------- | ------------------------------------------ |
-| `meta.subtasks`             | `pd.DataFrame \| None` | DataFrame mapping subtask names to indices |
-| `features["subtask_index"]` | `dict`                 | Feature spec for subtask_index if present  |
-
-### Sample Keys
-
-When subtasks are available, each sample includes:
-
-| Key             | Type           | Description                          |
-| --------------- | -------------- | ------------------------------------ |
-| `subtask_index` | `torch.Tensor` | Integer index of the current subtask |
-| `subtask`       | `str`          | Natural language subtask description |
-
-## Related Resources
-
-- [SARM Paper](https://arxiv.org/pdf/2509.25358) - Stage-Aware Reward Modeling for Long Horizon Robot Manipulation
-- [LeRobot Annotate Space](https://huggingface.co/spaces/lerobot/annotate) - Interactive annotation tool
-- [LeRobotDataset v3.0](./lerobot-dataset-v3) - Dataset format documentation

docs/source/earthrover_mini_plus.mdx

@@ -194,7 +194,7 @@ lerobot-record \
     --dataset.single_task="Navigate around obstacles" \
     --dataset.streaming_encoding=true \
     --dataset.encoder_threads=2 \
-    # --dataset.vcodec=auto \
+    # --dataset.camera_encoder.vcodec=auto \
     --display_data=true
 ```
 

docs/source/eo1.mdx

@@ -0,0 +1,168 @@
+# EO-1
+
+EO-1 is a **Vision-Language-Action policy for robot control**. The LeRobot implementation integrates EO-1 with the standard LeRobot training, evaluation, processor interface.
+
+## Model Overview
+
+EO-1 uses a Qwen2.5-VL backbone for vision-language understanding and adds a continuous flow-matching action head for robot control. The policy formats each robot-control sample as a multimodal conversation: camera images are passed to Qwen2.5-VL, the robot state is represented with EO-1 state tokens, and the future action chunk is represented with EO-1 action tokens.
+
+<img
+  src="https://huggingface.co/datasets/HaomingSong/lerobot-documentation-images/resolve/main/lerobot/eo_pipeline.png"
+  alt="An overview of EO-1"
+  width="85%"
+/>
+
+During training, EO-1 learns to denoise continuous action chunks at the action-token positions. During inference, it samples an action chunk, returns continuous actions, and executes `n_action_steps` from the chunk before sampling again.
+
+### What the LeRobot Integration Covers
+
+- Standard `policy.type=eo1` configuration through LeRobot
+- Qwen2.5-VL image and text preprocessing through policy processors
+- Continuous flow-matching action prediction
+- Checkpoint save/load through LeRobot policy APIs
+- Training with `lerobot-train` and evaluation with `lerobot-eval`
+
+The broader EO-1 project also includes interleaved vision-text-action pretraining and multimodal reasoning workflows. This page focuses on the LeRobot robot-control policy path.
+
+## Installation Requirements
+
+1. Install LeRobot by following the [Installation Guide](./installation).
+2. Install EO-1 dependencies by running:
+
+   ```bash
+   pip install -e ".[eo1]"
+   ```
+
+3. If you want to train or evaluate on LIBERO, install the LIBERO dependencies too:
+
+   ```bash
+   pip install -e ".[eo1,libero]"
+   ```
+
+EO-1 can use the standard PyTorch scaled-dot-product attention backend through `policy.attn_implementation=sdpa`. If your environment has a compatible `flash_attn` installation, you can request `policy.attn_implementation=flash_attention_2`.
+
+## Data Requirements
+
+EO-1 expects a LeRobot dataset with:
+
+- At least one visual observation, for example `observation.images.image`
+- `observation.state`
+- `action`
+- A language task instruction through the dataset `task` field
+
+If your dataset uses different observation names, use `rename_map` to align them with the names expected by your training or evaluation setup.
+
+## Usage
+
+To use EO-1 in a LeRobot configuration, specify the policy type as:
+
+```python
+policy.type=eo1
+```
+
+By default, a new EO-1 policy initializes its backbone from:
+
+```python
+policy.vlm_base=Qwen/Qwen2.5-VL-3B-Instruct
+```
+
+Once a LeRobot-format EO-1 checkpoint is available, load it with:
+
+```python
+policy.path=your-org/your-eo1-checkpoint
+```
+
+## Training
+
+### Training Command Example
+
+```bash
+lerobot-train \
+  --dataset.repo_id=your_org/your_dataset \
+  --policy.type=eo1 \
+  --policy.vlm_base=Qwen/Qwen2.5-VL-3B-Instruct \
+  --policy.dtype=bfloat16 \
+  --policy.attn_implementation=sdpa \
+  --policy.gradient_checkpointing=false \
+  --output_dir=./outputs/eo1_training \
+  --job_name=eo1_training \
+  --steps=300000 \
+  --batch_size=16 \
+  --policy.device=cuda
+```
+
+### Key Training Parameters
+
+| Parameter                              | Default                       | Description                                                             |
+| -------------------------------------- | ----------------------------- | ----------------------------------------------------------------------- |
+| `policy.vlm_base`                      | `Qwen/Qwen2.5-VL-3B-Instruct` | Qwen2.5-VL checkpoint used to initialize a new policy                   |
+| `policy.dtype`                         | `auto`                        | Backbone dtype request: `auto`, `bfloat16`, or `float32`                |
+| `policy.attn_implementation`           | `None`                        | Optional Qwen attention backend, such as `sdpa`                         |
+| `policy.gradient_checkpointing`        | `false`                       | Reduces memory usage during training                                    |
+| `policy.chunk_size`                    | `8`                           | Number of future actions predicted per chunk                            |
+| `policy.n_action_steps`                | `8`                           | Number of actions consumed from a sampled chunk                         |
+| `policy.num_denoise_steps`             | `10`                          | Number of flow-matching denoising steps used during sampling            |
+| `policy.max_state_dim`                 | `32`                          | State padding dimension                                                 |
+| `policy.max_action_dim`                | `32`                          | Action padding dimension                                                |
+| `policy.force_fp32_autocast`           | `true`                        | Keeps the flow head in fp32 even when the backbone uses mixed precision |
+| `policy.supervise_padding_action_dims` | `true`                        | Controls whether padded action dimensions are supervised                |
+| `policy.supervise_padding_actions`     | `true`                        | Controls whether padded future action rows are supervised               |
+
+## Evaluation
+
+EO-1 can be evaluated through `lerobot-eval` once you have a LeRobot-format checkpoint:
+
+```bash
+lerobot-eval \
+  --policy.path=your-org/your-eo1-checkpoint \
+  --env.type=libero \
+  --env.task=libero_object \
+  --eval.batch_size=1 \
+  --eval.n_episodes=20
+```
+
+For datasets or environments whose camera names differ from the checkpoint configuration, pass a `rename_map`:
+
+```bash
+lerobot-eval \
+  --policy.path=your-org/your-eo1-checkpoint \
+  --env.type=libero \
+  --env.task=libero_object \
+  --rename_map='{"observation.images.image2":"observation.images.wrist_image"}'
+```
+
+## Configuration Notes
+
+### Image Processing
+
+EO-1 uses the Qwen2.5-VL processor. The `policy.image_min_pixels` and `policy.image_max_pixels` settings control the image resizing bounds before the visual tokens are passed into the backbone.
+
+### State and Action Dimensions
+
+The policy pads state and action vectors to `policy.max_state_dim` and `policy.max_action_dim` before the EO-1 flow head. Predictions are cropped back to the original action dimension before being returned by the policy.
+
+### Attention Backend
+
+Use `policy.attn_implementation=sdpa` for a portable setup. Use `flash_attention_2` only when `flash_attn` is installed and compatible with your environment.
+
+## References
+
+- [EO-1 project](https://github.com/EO-Robotics/EO1)
+- [EO-1 paper](https://arxiv.org/abs/2508.21112)
+- [Qwen2.5-VL-3B-Instruct](https://huggingface.co/Qwen/Qwen2.5-VL-3B-Instruct)
+
+## Citation
+
+```bibtex
+@article{eo1,
+  title={EO-1: Interleaved Vision-Text-Action Pretraining for General Robot Control},
+  author={Delin Qu and Haoming Song and Qizhi Chen and Zhaoqing Chen and Xianqiang Gao and Xinyi Ye and Qi Lv and Modi Shi and Guanghui Ren and Cheng Ruan and Maoqing Yao and Haoran Yang and Jiacheng Bao and Bin Zhao and Dong Wang},
+  journal={arXiv preprint},
+  year={2025},
+  url={https://arxiv.org/abs/2508.21112}
+}
+```
+
+## License
+
+This LeRobot integration follows the **Apache 2.0 License** used by LeRobot. Check the upstream EO-1 model and dataset pages for the licenses of released EO-1 checkpoints and data.

docs/source/groot.mdx

@@ -105,10 +105,12 @@ These results demonstrate GR00T's strong generalization capabilities across dive
 
 ### Evaluate in your hardware setup
 
-Once you have trained your model using your parameters you can run inference in your downstream task. Follow the instructions in [Imitation Learning for Robots](./il_robots). For example:
+Once you have trained your model using your parameters you can run inference in your downstream task. Follow the instructions in [Policy Deployment (lerobot-rollout)](./inference). For example:
 
 ```bash
-lerobot-record \
+lerobot-rollout\
+  --strategy.type=sentry \
+  --strategy.upload_every_n_episodes=5 \
   --robot.type=bi_so_follower \
   --robot.left_arm_port=/dev/ttyACM1 \
   --robot.right_arm_port=/dev/ttyACM0 \
@@ -119,14 +121,12 @@ lerobot-record \
   }' \
   --display_data=true \
   --dataset.repo_id=<user>/eval_groot-bimanual  \
-  --dataset.num_episodes=10 \
   --dataset.single_task="Grab and handover the red cube to the other arm" \
   --dataset.streaming_encoding=true \
   --dataset.encoder_threads=2 \
-  # --dataset.vcodec=auto \
+  # --dataset.camera_encoder.vcodec=auto \
   --policy.path=<user>/groot-bimanual \ # your trained model
-  --dataset.episode_time_s=30 \
-  --dataset.reset_time_s=10
+  --duration=600
 ```
 
 ## License

docs/source/hardware_guide.mdx

@@ -0,0 +1,98 @@
+# Compute HW Guide for LeRobot Training
+
+Rough sizing for training a LeRobot policy: how much VRAM each policy needs, what training time looks like, and where to run when local hardware isn't enough.
+
+The numbers below are **indicative** — order-of-magnitude figures for picking hardware, not exact predictions. Throughput depends heavily on dataset I/O, image resolution, batch size, and number of GPUs.
+
+## Memory by policy group
+
+Policies cluster by backbone size; the groupings below give a single VRAM envelope per group instead of repeating numbers per policy. Memory scales roughly linearly with batch size; AdamW (the LeRobot default) carries optimizer state that adds ~30–100% over a forward+backward pass alone.
+
+| Group      | Policies                                    | Peak VRAM (BS 8, AdamW) | Suitable starter GPUs             |
+| ---------- | ------------------------------------------- | ----------------------: | --------------------------------- |
+| Light BC   | `act`, `vqbet`, `tdmpc`                     |                  ~2–6GB | Laptop GPU (RTX 3060), L4, A10G   |
+| Diffusion  | `diffusion`, `multi_task_dit`               |                 ~8–14GB | RTX 4070+ / L4 / A10G             |
+| Small VLA  | `smolvla`                                   |                ~10–16GB | RTX 4080+ / L4 / A10G             |
+| Large VLA  | `pi0`, `pi0_fast`, `pi05`, `xvla`, `wall_x` |                ~24–40GB | A100 40 GB+ (24 GB tight at BS 1) |
+| Multimodal | `groot`, `eo1`                              |                ~24–40GB | A100 40 GB+                       |
+| RL         | `sac`                                       |             config-dep. | See [HIL-SERL guide](./hilserl)   |
+
+Memory-bound? Drop the batch size (~linear), use gradient accumulation to recover effective batch, or for SmolVLA leave `freeze_vision_encoder=True`.
+
+## Training time
+
+Robotics imitation learning typically converges in **5–10 epochs over the dataset**, not hundreds of thousands of raw steps. Once you know your epoch count, wall-clock is essentially:
+
+```text
+total_frames    = sum of frames over all episodes      # 50 ep × 30 fps × 30 s ≈ 45,000
+steps_per_epoch = ceil(total_frames / (num_gpus × batch_size))
+total_steps     = epochs × steps_per_epoch
+wall_clock      ≈ total_steps × per_step_time
+```
+
+Per-step time depends on the policy and the GPU. The numbers in the table below are anchors — pick the row closest to your setup and scale linearly with `total_steps` if you train longer or shorter.
+
+### Common scenarios
+
+Indicative wall-clock for **5 epochs on a ~50-episode dataset (~45k frames at 30 fps × 30 s)**, default optimizer (AdamW), 640×480 images:
+
+| Setup                                | Policy         | Batch | Wall-clock |
+| ------------------------------------ | -------------- | ----- | ---------: |
+| Single RTX 4090 / RTX 3090 (24 GB)   | `act`          | 8     |  ~30–60min |
+| Single RTX 4090 / RTX 3090 (24 GB)   | `diffusion`    | 8     |      ~2–4h |
+| Single L4 / A10G (24 GB)             | `act`          | 8     |      ~1–2h |
+| Single L4 / A10G (24 GB)             | `smolvla`      | 4     |      ~3–6h |
+| Single A100 40 GB                    | `smolvla`      | 16    |      ~1–2h |
+| Single A100 40 GB                    | `pi0` / `pi05` | 4     |      ~4–8h |
+| 4× H100 80 GB cluster (`accelerate`) | `diffusion`    | 32    |  ~30–60min |
+| 4× H100 80 GB cluster (`accelerate`) | `smolvla`      | 32    |      ~1–2h |
+| Apple Silicon M1/M2/M3 Max (MPS)     | `act`          | 4     |     ~6–14h |
+
+These are order-of-magnitude figures. Real runs deviate by ±50% depending on image resolution, dataset I/O, dataloader threading, and exact GPU SKU. They are useful as "is this run going to take an hour or a day?" intuition, not as SLAs.
+
+### Multi-GPU matters a lot
+
+`accelerate launch --num_processes=N` is the easiest way to cut training time. Each optimizer step processes `N × batch_size` samples in roughly the same wall-clock as a single-GPU step, so 4 GPUs ≈ 4× speedup for compute-bound runs. See the [Multi GPU training](./multi_gpu_training) guide for the full setup.
+
+Reference data points on a 4×H100 80 GB cluster (`accelerate launch --num_processes=4`), 5000 steps, batch 32, AdamW, dataset [`imstevenpmwork/super_poulain_draft`](https://huggingface.co/datasets/imstevenpmwork/super_poulain_draft) (~50 episodes, ~640×480 images):
+
+| Policy      | Wall-clock | `update_s` | `dataloading_s` | GPU util | Notable flags                                                                                                                  |
+| ----------- | ---------- | ---------: | --------------: | -------- | ------------------------------------------------------------------------------------------------------------------------------ |
+| `diffusion` | 16m 17s    |      0.167 |           0.015 | ~90%     | defaults (training from scratch)                                                                                               |
+| `smolvla`   | 27m 49s    |      0.312 |           0.011 | ~80%     | `--policy.path=lerobot/smolvla_base`, `freeze_vision_encoder=false`, `train_expert_only=false`                                 |
+| `pi05`      | 3h 41m     |      2.548 |           0.014 | ~95%     | `--policy.pretrained_path=lerobot/pi05_base`, `gradient_checkpointing=true`, `dtype=bfloat16`, vision encoder + expert trained |
+
+The `dataloading_s` vs. `update_s` ratio is the diagnostic that matters: when `dataloading_s` approaches `update_s`, more GPUs stop helping — your dataloader is the bottleneck and you should look at `--num_workers`, image resolution, and disk speed before adding compute.
+
+### Schedule and checkpoints
+
+If you shorten training (e.g. 5k–10k steps on a small dataset), also shorten the LR schedule with `--policy.scheduler_decay_steps≈--steps`. Otherwise the LR stays near its peak and never decays. Same for `--save_freq`.
+
+## Where to run
+
+VRAM is the first filter. Within a tier, pick by budget and availability — the `$`–`$$$$` columns are relative; check current pricing on the provider you actually use.
+
+| Class                      | VRAM  | Tier   | Comfortable for                                             |
+| -------------------------- | ----- | ------ | ----------------------------------------------------------- |
+| RTX 3090 / 4090 (consumer) | 24 GB | `$`    | Light BC, Diffusion, SmolVLA. Tight for VLAs at batch 1.    |
+| L4 / A10G (cloud)          | 24 GB | `$–$$` | Same envelope; common on Google Cloud, RunPod, AWS `g5/g6`. |
+| A100 40 GB                 | 40 GB | `$$$`  | Any policy at reasonable batch sizes.                       |
+| A100 80 GB / H100 80 GB    | 80 GB | `$$$$` | Multi-GPU clusters; large batches for VLAs.                 |
+| **CPU only**               | —     | —      | Don't train. Use Colab or rent a GPU.                       |
+
+### Hugging Face Jobs
+
+[Hugging Face Jobs](https://huggingface.co/docs/hub/jobs) lets you run training on managed HF infrastructure, billed by the second. The repo publishes a ready-to-use image: **`huggingface/lerobot-gpu:latest`**, rebuilt **every night at 02:00 UTC from `main`** ([`docker_publish.yml`](https://github.com/huggingface/lerobot/blob/main/.github/workflows/docker_publish.yml)) — so it tracks the current state of the repo, not a tagged release.
+
+```bash
+hf jobs run --flavor a10g-large huggingface/lerobot-gpu:latest \
+  bash -c "nvidia-smi && lerobot-train \
+    --policy.type=act --dataset.repo_id=<USER>/<DATASET> \
+    --policy.repo_id=<USER>/act_<task> --batch_size=8 --steps=50000"
+```
+
+Notes:
+
+- The leading `nvidia-smi` is a quick sanity check that CUDA is visible inside the container — useful to fail fast if the flavor or driver mismatched.
+- The default Job timeout is 30 minutes; pass `--timeout 4h` (or longer) for real training.
+- `--flavor` maps onto the table above: `t4-small`/`t4-medium` (T4, ACT only), `l4x1`/`l4x4` (L4 24 GB), `a10g-small/large/largex2/largex4` (A10G 24 GB scaled out), `a100-large` (A100). For the current full catalogue + pricing see [https://huggingface.co/docs/hub/jobs](https://huggingface.co/docs/hub/jobs).

docs/source/hilserl.mdx

@@ -62,7 +62,7 @@ pip install -e ".[hilserl]"
 
 ### Understanding Configuration
 
-The training process begins with proper configuration for the HILSerl environment. The main configuration class is `GymManipulatorConfig` in `lerobot/rl/gym_manipulator.py`, which contains nested `HILSerlRobotEnvConfig` and `DatasetConfig`. The configuration is organized into focused, nested sub-configs:
+The training process begins with proper configuration for the HILSERl environment. The main configuration class is `GymManipulatorConfig` in `lerobot/rl/gym_manipulator.py`, which contains nested `HILSerlRobotEnvConfig` (defined in `lerobot/envs/configs.py`) and `DatasetConfig`. The configuration is organized into focused, nested sub-configs:
 
 <!-- prettier-ignore-start -->
 ```python
@@ -95,6 +95,7 @@ class HILSerlProcessorConfig:
 class ObservationConfig:
     add_joint_velocity_to_observation: bool = False    # Add joint velocities to state
     add_current_to_observation: bool = False    # Add motor currents to state
+    add_ee_pose_to_observation: bool = False    # Add end-effector pose to state
     display_cameras: bool = False    # Display camera feeds during execution
 
 class ImagePreprocessingConfig:
@@ -326,14 +327,22 @@ lerobot-find-joint-limits \
    Max joint positions [-20.0, -20.0, -20.0, -20.0, -20.0, -20.0]
    Min joint positions [50.0, 50.0, 50.0, 50.0, 50.0, 50.0]
    ```
-3. Use these values in the configuration of your teleoperation device (TeleoperatorConfig) under the `end_effector_bounds` field
+3. Use these values in your environment configuration under `env.processor.inverse_kinematics.end_effector_bounds` (see `InverseKinematicsConfig` in `lerobot/envs/configs.py`)
 
 **Example Configuration**
 
 ```json
-"end_effector_bounds": {
-    "max": [0.24, 0.20, 0.10],
-    "min": [0.16, -0.08, 0.03]
+{
+  "env": {
+    "processor": {
+      "inverse_kinematics": {
+        "end_effector_bounds": {
+          "max": [0.24, 0.2, 0.1],
+          "min": [0.16, -0.08, 0.03]
+        }
+      }
+    }
+  }
 }
 ```
 
@@ -404,30 +413,24 @@ We support using a gamepad or a keyboard or the leader arm of the robot.
 
 HIL-Serl learns actions in the end-effector space of the robot. Therefore, the teleoperation will control the end-effector's x,y,z displacements.
 
-For that we need to define a version of the robot that takes actions in the end-effector space. Check the robot class `SO100FollowerEndEffector` and its configuration `SO100FollowerEndEffectorConfig` for the default parameters related to the end-effector space.
+The end-effector transformation is applied by the processor pipeline (`InverseKinematicsRLStep`, `EEBoundsAndSafety`, `EEReferenceAndDelta`, `GripperVelocityToJoint`) configured under `env.processor.inverse_kinematics` (`InverseKinematicsConfig`) and `env.processor.gripper` / `env.processor.max_gripper_pos`. The defaults related to the end-effector space are:
 
 <!-- prettier-ignore-start -->
 ```python
-class SO100FollowerEndEffectorConfig(SO100FollowerConfig):
-    """Configuration for the SO100FollowerEndEffector robot."""
+class InverseKinematicsConfig:
+    """Configuration for inverse kinematics processing."""
 
-    # Default bounds for the end-effector position (in meters)
-    end_effector_bounds: dict[str, list[float]] = field( # bounds for the end-effector in x,y,z direction
-        default_factory=lambda: {
-            "min": [-1.0, -1.0, -1.0],  # min x, y, z
-            "max": [1.0, 1.0, 1.0],  # max x, y, z
-        }
-    )
+    urdf_path: str | None = None
+    target_frame_name: str | None = None
+    # bounds for the end-effector in x,y,z direction
+    end_effector_bounds: dict[str, list[float]] | None = None
+    # maximum step size for the end-effector in x,y,z direction
+    end_effector_step_sizes: dict[str, float] | None = None
 
-    max_gripper_pos: float = 50 # maximum gripper position that the gripper will be open at
-
-    end_effector_step_sizes: dict[str, float] = field( # maximum step size for the end-effector in x,y,z direction
-        default_factory=lambda: {
-            "x": 0.02,
-            "y": 0.02,
-            "z": 0.02,
-        }
-    )
+class HILSerlProcessorConfig:
+    ...
+    # maximum gripper position that the gripper will be open at
+    max_gripper_pos: float | None = 100.0
 ```
 <!-- prettier-ignore-end -->
 
@@ -606,11 +609,11 @@ This guide explains how to train a reward classifier for human-in-the-loop reinf
 
 **Note**: Training a reward classifier is optional. You can start the first round of RL experiments by annotating the success manually with your gamepad or keyboard device.
 
-The reward classifier implementation in `modeling_classifier.py` uses a pretrained vision model to process the images. It can output either a single value for binary rewards to predict success/fail cases or multiple values for multi-class settings.
+The reward classifier implementation in `lerobot/rewards/classifier/modeling_classifier.py` uses a pretrained vision model to process the images. It can output either a single value for binary rewards to predict success/fail cases or multiple values for multi-class settings.
 
 **Collecting a Dataset for the reward classifier**
 
-Before training, you need to collect a dataset with labeled examples. The `record_dataset` function in `gym_manipulator.py` enables the process of collecting a dataset of observations, actions, and rewards.
+Before training, you need to collect a dataset with labeled examples. Setting `mode: "record"` in your config and running `gym_manipulator.py` enables the process of collecting a dataset of observations, actions, and rewards.
 
 To collect a dataset, you need to modify some parameters in the environment configuration based on HILSerlRobotEnvConfig.
 
@@ -658,7 +661,7 @@ Example configuration section for data collection:
   },
   "dataset": {
     "repo_id": "hf_username/dataset_name",
-    "dataset_root": "data/your_dataset",
+    "root": "data/your_dataset",
     "task": "reward_classifier_task",
     "num_episodes_to_record": 20,
     "replay_episode": null,
@@ -671,7 +674,7 @@ Example configuration section for data collection:
 
 **Reward Classifier Configuration**
 
-The reward classifier is configured using `configuration_classifier.py`. Here are the key parameters:
+The reward classifier is configured using `lerobot/rewards/classifier/configuration_classifier.py`. Here are the key parameters:
 
 - **model_name**: Base model architecture (e.g., we mainly use `"helper2424/resnet10"`)
 - **model_type**: `"cnn"` or `"transformer"`
@@ -689,7 +692,7 @@ Example configuration for training the [reward classifier](https://huggingface.c
     "repo_id": "hf_username/dataset_name",
     "root": null
   },
-  "policy": {
+  "reward_model": {
     "type": "reward_classifier",
     "model_name": "helper2424/resnet10",
     "model_type": "cnn",
@@ -699,7 +702,6 @@ Example configuration for training the [reward classifier](https://huggingface.c
     "dropout_rate": 0.1,
     "learning_rate": 1e-4,
     "device": "cuda",
-    "use_amp": true,
     "input_features": {
       "observation.images.front": {
         "type": "VISUAL",
@@ -818,13 +820,14 @@ The LeRobot system uses a distributed actor-learner architecture for training. T
 
 **Configuration Setup**
 
-Create a training configuration file (example available [here](https://huggingface.co/datasets/lerobot/config_examples/resolve/main/rl/train_config.json)). The training config is based on the main `TrainRLServerPipelineConfig` class in `lerobot/configs/train.py`.
+Create a training configuration file (example available [here](https://huggingface.co/datasets/lerobot/config_examples/resolve/main/rl/train_config.json)). The training config is based on the main `TrainRLServerPipelineConfig` class in `lerobot/rl/train_rl.py`.
 
-1. Configure the policy settings (`type="sac"`, `device`, etc.)
-2. Set `dataset` to your cropped dataset
-3. Configure environment settings with crop parameters
-4. Check the other parameters related to SAC in [configuration_sac.py](https://github.com/huggingface/lerobot/blob/main/src/lerobot/policies/sac/configuration_sac.py#L79).
-5. Verify that the `policy` config is correct with the right `input_features` and `output_features` for your task.
+1. Configure the policy settings (`type="gaussian_actor"`, `device`, etc.)
+2. Configure the algorithm settings under the top-level `algorithm` block (`type="sac"`, learning rates, discount, etc., defined in `lerobot/rl/algorithms/sac/configuration_sac.py`).
+3. Set `dataset` to your cropped dataset
+4. Configure environment settings with crop parameters
+5. Check the other parameters related to the Gaussian Actor in [configuration_gaussian_actor.py](https://github.com/huggingface/lerobot/blob/main/src/lerobot/policies/gaussian_actor/configuration_gaussian_actor.py#L79).
+6. Verify that the `policy` config is correct with the right `input_features` and `output_features` for your task.
 
 **Starting the Learner**
 
@@ -926,7 +929,7 @@ The ideal behaviour is that your intervention rate should drop gradually during
 
 Some configuration values have a disproportionate impact on training stability and speed:
 
-- **`temperature_init`** (`policy.temperature_init`) – initial entropy temperature in SAC. Higher values encourage more exploration; lower values make the policy more deterministic early on. A good starting point is `1e-2`. We observed that setting it too high can make human interventions ineffective and slow down learning.
+- **`temperature_init`** (`algorithm.temperature_init`) – initial entropy temperature in SAC. Higher values encourage more exploration; lower values make the policy more deterministic early on. A good starting point is `1e-2`. We observed that setting it too high can make human interventions ineffective and slow down learning.
 - **`policy_parameters_push_frequency`** (`policy.actor_learner_config.policy_parameters_push_frequency`) – interval in _seconds_ between two weight pushes from the learner to the actor. The default is `4 s`. Decrease to **1-2 s** to provide fresher weights (at the cost of more network traffic); increase only if your connection is slow, as this will reduce sample efficiency.
 - **`storage_device`** (`policy.storage_device`) – device on which the learner keeps the policy parameters. If you have spare GPU memory, set this to `"cuda"` (instead of the default `"cpu"`). Keeping the weights on-GPU removes CPU→GPU transfer overhead and can significantly increase the number of learner updates per second.
 

docs/source/hope_jr.mdx

@@ -232,7 +232,7 @@ lerobot-record \
     --dataset.private=true \
     --dataset.streaming_encoding=true \
     --dataset.encoder_threads=2 \
-    # --dataset.vcodec=auto \
+    # --dataset.camera_encoder.vcodec=auto \
     --display_data=true
 ```
 
@@ -278,6 +278,6 @@ lerobot-record \
   --dataset.num_episodes=10 \
   --dataset.streaming_encoding=true \
   --dataset.encoder_threads=2 \
-  # --dataset.vcodec=auto \
+  # --dataset.camera_encoder.vcodec=auto \
   --policy.path=outputs/train/hopejr_hand/checkpoints/last/pretrained_model
 ```

docs/source/il_robots.mdx

@@ -68,13 +68,13 @@ from lerobot.teleoperators.so_leader import SO101Leader, SO101LeaderConfig
 from lerobot.robots.so_follower import SO101Follower, SO101FollowerConfig
 
 robot_config = SO101FollowerConfig(
-    port="/dev/tty.usbmodem58760431541",
-    id="my_red_robot_arm",
+    port="/dev/tty.usbmodem5AB90687491",
+    id="my_follower_arm",
 )
 
 teleop_config = SO101LeaderConfig(
-    port="/dev/tty.usbmodem58760431551",
-    id="my_blue_leader_arm",
+    port="/dev/tty.usbmodem5AB90689011",
+    id="my_leader_arm",
 )
 
 robot = SO101Follower(robot_config)
@@ -108,13 +108,13 @@ With `rerun`, you can teleoperate again while simultaneously visualizing the cam
 <hfoption id="Command">
 ```bash
 lerobot-teleoperate \
-    --robot.type=koch_follower \
-    --robot.port=/dev/tty.usbmodem58760431541 \
-    --robot.id=my_awesome_follower_arm \
-    --robot.cameras="{ front: {type: opencv, index_or_path: 0, width: 1920, height: 1080, fps: 30}}" \
-    --teleop.type=koch_leader \
-    --teleop.port=/dev/tty.usbmodem58760431551 \
-    --teleop.id=my_awesome_leader_arm \
+    --robot.type=so101_follower \
+    --robot.port=/dev/tty.usbmodem5AB90687491 \
+    --robot.id=my_follower_arm \
+    --robot.cameras="{front: {type: opencv, index_or_path: 0, width: 640, height: 480, fps: 30}}" \
+    --teleop.type=so101_leader \
+    --teleop.port=/dev/tty.usbmodem5AB90689011 \
+    --teleop.id=my_leader_arm \
     --display_data=true
 ```
 </hfoption>
@@ -122,34 +122,48 @@ lerobot-teleoperate \
 
 <!-- prettier-ignore-start -->
 ```python
+import time
+from lerobot.teleoperators.so_leader import SO101Leader, SO101LeaderConfig
+from lerobot.robots.so_follower import SO101Follower, SO101FollowerConfig
 from lerobot.cameras.opencv import OpenCVCameraConfig
-from lerobot.teleoperators.koch_leader import KochLeader, KochLeaderConfig
-from lerobot.robots.koch_follower import KochFollower, KochFollowerConfig
+from lerobot.utils.visualization_utils import init_rerun, log_rerun_data, shutdown_rerun
 
-camera_config = {
-    "front": OpenCVCameraConfig(index_or_path=0, width=1920, height=1080, fps=30)
-}
-
-robot_config = KochFollowerConfig(
-    port="/dev/tty.usbmodem585A0076841",
-    id="my_red_robot_arm",
-    cameras=camera_config
+robot_config = SO101FollowerConfig(
+    port="/dev/tty.usbmodem5AB90687491",
+    id="my_follower_arm",
+    cameras={
+        "wrist": OpenCVCameraConfig(index_or_path=0, width=640, height=480, fps=30),
+        "top": OpenCVCameraConfig(index_or_path=1, width=640, height=480, fps=30)
+    }
 )
 
-teleop_config = KochLeaderConfig(
-    port="/dev/tty.usbmodem58760431551",
-    id="my_blue_leader_arm",
+teleop_config = SO101LeaderConfig(
+    port="/dev/tty.usbmodem5AB90689011",
+    id="my_leader_arm",
 )
 
-robot = KochFollower(robot_config)
-teleop_device = KochLeader(teleop_config)
+init_rerun(session_name="teleoperation")
+
+robot = SO101Follower(robot_config)
+teleop_device = SO101Leader(teleop_config)
 robot.connect()
 teleop_device.connect()
 
+TARGET_HZ = 30
+TIME_PER_FRAME = 1.0 / TARGET_HZ
+
 while True:
+    start_time = time.perf_counter()
+
     observation = robot.get_observation()
     action = teleop_device.get_action()
     robot.send_action(action)
+    log_rerun_data(observation=observation, action=action)
+
+    elapsed_time = time.perf_counter() - start_time
+    sleep_time = TIME_PER_FRAME - elapsed_time
+    if sleep_time > 0:
+        time.sleep(sleep_time)
 ```
 <!-- prettier-ignore-end -->
 
@@ -193,7 +207,7 @@ lerobot-record \
     --dataset.num_episodes=5 \
     --dataset.single_task="Grab the black cube" \
     --dataset.streaming_encoding=true \
-    # --dataset.vcodec=auto \
+    # --dataset.camera_encoder.vcodec=auto \
     --dataset.encoder_threads=2
 ```
 </hfoption>
@@ -202,10 +216,11 @@ lerobot-record \
 <!-- prettier-ignore-start -->
 ```python
 from lerobot.cameras.opencv import OpenCVCameraConfig
-from lerobot.datasets import LeRobotDataset
+from lerobot.datasets.lerobot_dataset import LeRobotDataset
 from lerobot.utils.feature_utils import hw_to_dataset_features
-from lerobot.robots.so_follower import SO100Follower, SO100FollowerConfig
-from lerobot.teleoperators.so_leader import SO100Leader, SO100LeaderConfig
+from lerobot.robots.so_follower import SO101Follower, SO101FollowerConfig
+from lerobot.teleoperators.so_leader.config_so_leader import SO101LeaderConfig
+from lerobot.teleoperators.so_leader.so_leader import SO101Leader
 from lerobot.common.control_utils import init_keyboard_listener
 from lerobot.utils.utils import log_say
 from lerobot.utils.visualization_utils import init_rerun
@@ -218,71 +233,56 @@ EPISODE_TIME_SEC = 60
 RESET_TIME_SEC = 10
 TASK_DESCRIPTION = "My task description"
 
-# Create robot configuration
-robot_config = SO100FollowerConfig(
-    id="my_awesome_follower_arm",
-    cameras={
-        "front": OpenCVCameraConfig(index_or_path=0, width=640, height=480, fps=FPS) # Optional: fourcc="MJPG" for troubleshooting OpenCV async error.
-    },
-    port="/dev/tty.usbmodem58760434471",
-)
-
-teleop_config = SO100LeaderConfig(
-    id="my_awesome_leader_arm",
-    port="/dev/tty.usbmodem585A0077581",
-)
-
-# Initialize the robot and teleoperator
-robot = SO100Follower(robot_config)
-teleop = SO100Leader(teleop_config)
-
-# Configure the dataset features
-action_features = hw_to_dataset_features(robot.action_features, "action")
-obs_features = hw_to_dataset_features(robot.observation_features, "observation")
-dataset_features = {**action_features, **obs_features}
-
-# Create the dataset
-dataset = LeRobotDataset.create(
-    repo_id="<hf_username>/<dataset_repo_id>",
-    fps=FPS,
-    features=dataset_features,
-    robot_type=robot.name,
-    use_videos=True,
-    image_writer_threads=4,
-)
-
-# Initialize the keyboard listener and rerun visualization
-_, events = init_keyboard_listener()
-init_rerun(session_name="recording")
-
-# Connect the robot and teleoperator
-robot.connect()
-teleop.connect()
-
-# Create the required processors
-teleop_action_processor, robot_action_processor, robot_observation_processor = make_default_processors()
-
-episode_idx = 0
-while episode_idx < NUM_EPISODES and not events["stop_recording"]:
-    log_say(f"Recording episode {episode_idx + 1} of {NUM_EPISODES}")
-
-    record_loop(
-        robot=robot,
-        events=events,
-        fps=FPS,
-        teleop_action_processor=teleop_action_processor,
-        robot_action_processor=robot_action_processor,
-        robot_observation_processor=robot_observation_processor,
-        teleop=teleop,
-        dataset=dataset,
-        control_time_s=EPISODE_TIME_SEC,
-        single_task=TASK_DESCRIPTION,
-        display_data=True,
+def main():
+    # Create robot configuration
+    robot_config = SO101FollowerConfig(
+        port="/dev/tty.usbmodem5AB90687491",
+        id="my_follower_arm",
+        cameras={
+            "wrist": OpenCVCameraConfig(index_or_path=0, width=640, height=480, fps=30),
+            "top": OpenCVCameraConfig(index_or_path=1, width=640, height=480, fps=30)
+        }
     )
 
-    # Reset the environment if not stopping or re-recording
-    if not events["stop_recording"] and (episode_idx < NUM_EPISODES - 1 or events["rerecord_episode"]):
-        log_say("Reset the environment")
+    teleop_config = SO101LeaderConfig(
+        port="/dev/tty.usbmodem5AB90689011",
+        id="my_leader_arm",
+    )
+
+    # Initialize the robot and teleoperator
+    robot = SO101Follower(robot_config)
+    teleop = SO101Leader(teleop_config)
+
+    # Configure the dataset features
+    action_features = hw_to_dataset_features(robot.action_features, "action")
+    obs_features = hw_to_dataset_features(robot.observation_features, "observation")
+    dataset_features = {**action_features, **obs_features}
+
+    # Create the dataset
+    dataset = LeRobotDataset.create(
+        repo_id="<hf_username>/<dataset_repo_id>",
+        fps=FPS,
+        features=dataset_features,
+        robot_type=robot.name,
+        use_videos=True,
+        image_writer_threads=4,
+    )
+
+    # Initialize the keyboard listener and rerun visualization
+    _, events = init_keyboard_listener()
+    init_rerun(session_name="recording")
+
+    # Connect the robot and teleoperator
+    robot.connect()
+    teleop.connect()
+
+    # Create the required processors
+    teleop_action_processor, robot_action_processor, robot_observation_processor = make_default_processors()
+
+    episode_idx = 0
+    while episode_idx < NUM_EPISODES and not events["stop_recording"]:
+        log_say(f"Recording episode {episode_idx + 1} of {NUM_EPISODES}")
+
         record_loop(
             robot=robot,
             events=events,
@@ -291,26 +291,50 @@ while episode_idx < NUM_EPISODES and not events["stop_recording"]:
             robot_action_processor=robot_action_processor,
             robot_observation_processor=robot_observation_processor,
             teleop=teleop,
-            control_time_s=RESET_TIME_SEC,
+            dataset=dataset,
+            control_time_s=EPISODE_TIME_SEC,
             single_task=TASK_DESCRIPTION,
             display_data=True,
         )
 
-    if events["rerecord_episode"]:
-        log_say("Re-recording episode")
-        events["rerecord_episode"] = False
-        events["exit_early"] = False
-        dataset.clear_episode_buffer()
-        continue
+        # Reset the environment if not stopping or re-recording
+        if not events["stop_recording"] and (episode_idx < NUM_EPISODES - 1 or events["rerecord_episode"]):
+            log_say("Reset the environment")
+            record_loop(
+                robot=robot,
+                events=events,
+                fps=FPS,
+                teleop_action_processor=teleop_action_processor,
+                robot_action_processor=robot_action_processor,
+                robot_observation_processor=robot_observation_processor,
+                teleop=teleop,
+                control_time_s=RESET_TIME_SEC,
+                single_task=TASK_DESCRIPTION,
+                display_data=True,
+            )
 
-    dataset.save_episode()
-    episode_idx += 1
+        if events["rerecord_episode"]:
+            log_say("Re-recording episode")
+            events["rerecord_episode"] = False
+            events["exit_early"] = False
+            dataset.clear_episode_buffer()
+            continue
 
-# Clean up
-log_say("Stop recording")
-robot.disconnect()
-teleop.disconnect()
-dataset.push_to_hub()
+        dataset.save_episode()
+        episode_idx += 1
+
+    # finalize dataset
+    log_say("Finalizing dataset...")
+    dataset.finalize()
+    # Clean up
+    log_say("Stop recording")
+    robot.disconnect()
+    teleop.disconnect()
+    dataset.push_to_hub()
+
+
+if __name__ == "__main__":
+    main()
 ```
 <!-- prettier-ignore-end -->
 
@@ -348,7 +372,7 @@ The `record` function provides a suite of tools for capturing and managing data
 ##### 2. Checkpointing and Resuming
 
 - Checkpoints are automatically created during recording.
-- If an issue occurs, you can resume by re-running the same command with `--resume=true`. When resuming a recording, `--dataset.num_episodes` must be set to the **number of additional episodes to be recorded**, and not to the targeted total number of episodes in the dataset !
+- If an issue occurs or you want to record additional episodes in the same dataset, you can resume by re-running the same command with `--resume=true`. When resuming a recording, `--dataset.num_episodes` must be set to the **number of additional episodes to be recorded**, and not to the targeted total number of episodes in the dataset! Make sure that you also set `--dataset.root="local_path"`, it's a local path to save the new part of the dataset and is required to resume.
 - To start recording from scratch, **manually delete** the dataset directory.
 
 ##### 3. Recording Parameters
@@ -422,7 +446,7 @@ from lerobot.utils.utils import log_say
 
 episode_idx = 0
 
-robot_config = SO100FollowerConfig(port="/dev/tty.usbmodem58760434471", id="my_awesome_follower_arm")
+robot_config = SO100FollowerConfig(port="/dev/tty.usbmodem5AB90687491", id="my_follower_arm")
 
 robot = SO100Follower(robot_config)
 robot.connect()
@@ -490,6 +514,83 @@ Additionally you can provide extra `tags` or specify a `license` for your model
 
 If your local computer doesn't have a powerful GPU you could utilize Google Colab to train your model by following the [ACT training notebook](./notebooks#training-act).
 
+#### Train using Hugging Face Jobs
+
+Hugging Face jobs let's you easily select hardware and run the training in the cloud. So if you don't have a powerful GPU or you need more VRAM or just want to train a model much faster use HF Jobs! It's pay as you go and you simply pay for each second of use, you can see the pricing and additional information [here](https://huggingface.co/docs/hub/jobs).
+
+To run the training use this command:
+
+<hfoptions id="train_with_hf_jobs">
+<hfoption id="Command">
+```bash
+hf jobs run \
+  --flavor a10g-small \
+  --timeout 4h \
+  --secrets HF_TOKEN \
+  huggingface/lerobot-gpu:latest \
+  -- \
+  python -m lerobot.scripts.lerobot_train \
+    --dataset.repo_id=username/dataset \
+    --policy.type=act \
+    --steps=5000 \
+    --batch_size=16 \
+    --policy.device=cuda \
+    --policy.repo_id=username/your_policy \
+    --log_freq=100
+```
+</hfoption>
+<hfoption id="API example">
+
+<!-- prettier-ignore-start -->
+```python
+from huggingface_hub import run_job, get_token
+
+run_name = "act_so101_hf_jobs"
+dataset_id = "username/dataset"
+user_hub_id = "username"
+
+command_args = [
+    "python", "-m", "lerobot.scripts.lerobot_train",
+    "--dataset.repo_id", dataset_id,
+    "--policy.type", "act",
+    "--steps", "5000",
+    "--batch_size", "16",
+    "--num_workers", "4",
+    "--policy.device", "cuda",
+    "--log_freq", "100",
+    "--save_freq", "1000",
+    "--save_checkpoint", "true",
+    "--wandb.enable", "false",
+    "--policy.repo_id", f"{user_hub_id}/{run_name}"
+]
+
+print(f"Submitting job '{run_name}' to Hugging Face Infrastructure...")
+
+job_info = run_job(
+    image="huggingface/lerobot-gpu:latest",
+    command=command_args,
+    flavor="a10g-small",
+    timeout="4h",
+    secrets={"HF_TOKEN": get_token()}
+)
+
+print("\n🚀 Job successfully launched!")
+print(f"🔹 Job ID: {job_info.id}")
+print(f"🔗 Live UI Dashboard & Logs: {job_info.url}")
+```
+<!-- prettier-ignore-end -->
+
+</hfoption>
+</hfoptions>
+
+You can modify the `--flavor` to use different hardware, for example: `t4-small`, `a100-large`, `h200`. Use `hf jobs hardware` to see the full list with pricing.
+Depending on the model you want to train and the hardware you selected you can also modify the `--batch_size` and `--number_of_workers`.
+For longer training sessions increase the timeout.
+
+Once the training is started you can go to [Jobs](https://huggingface.co/settings/jobs) and see if your jobs is running as well as all the outputs. Sometimes it takes a few minutes to schedule your job so be patient.
+
+After training the model will be pushed to hub and you can use it as any other model with LeRobot.
+
 #### Upload policy checkpoints
 
 Once training is done, upload the latest checkpoint with:

docs/source/installation.mdx

@@ -207,6 +207,56 @@ pip install 'lerobot[feetech]'        # Feetech motor support
 
 _Multiple extras can be combined (e.g., `.[core_scripts,pi,pusht]`). For a full list of available extras, refer to `pyproject.toml`._
 
+### PyTorch CUDA variant (Linux only)
+
+On Linux, the install path determines which CUDA wheel you get. macOS and Windows installs use the PyPI default (MPS / CPU / CUDA-Windows wheel respectively) and can skip this section.
+
+<!-- prettier-ignore-start -->
+
+<hfoptions id="cuda_variant">
+<hfoption id="uv-source">
+
+**Source install via `uv` (`uv sync` or `uv pip install -e .`)**
+
+`torch` and `torchvision` are pinned by the project to the **CUDA 12.8** PyTorch index (`https://download.pytorch.org/whl/cu128`, driver floor **570.86**) — covers Ampere/Ada/Hopper/Blackwell GPUs. No action needed for typical NVIDIA setups.
+
+To override for a different CUDA variant:
+
+```bash
+uv pip install --force-reinstall torch torchvision \
+    --index-url https://download.pytorch.org/whl/cu126   # older drivers; or cu130 for Blackwell on driver ≥ 580
+```
+
+</hfoption>
+<hfoption id="pip-conda">
+
+**Source install via `pip`/`conda`, or `pip install lerobot` from PyPI**
+
+PyPI default torch wheel is currently a cu130-bundled Linux wheel, driver floor **580.65**.
+
+To pick a specific CUDA variant:
+
+**Using `pip` or `conda`** — install torch first with an explicit index, then lerobot:
+
+```bash
+pip install --index-url https://download.pytorch.org/whl/cu128 torch torchvision
+pip install -e ".[all]"          # source
+# — or —
+pip install lerobot              # from PyPI
+```
+
+**Using `uv` to install from PyPI** — one-liner via `--torch-backend` (uv ≥ 0.6):
+
+```bash
+uv pip install --torch-backend cu128 lerobot
+```
+
+Supported values include `auto`, `cpu`, `cu126`, `cu128`, `cu129`, `cu130`, plus various `rocm*` and `xpu`. Swap as needed for your driver.
+
+</hfoption>
+</hfoptions>
+<!-- prettier-ignore-end -->
+
 ### Troubleshooting
 
 If you encounter build errors, you may need to install additional system dependencies: `cmake`, `build-essential`, and `ffmpeg libs`.

docs/source/language_and_recipes.mdx

@@ -0,0 +1,147 @@
+# Language columns and recipes
+
+Most LeRobot datasets ship with a single `task` string per episode — fine for
+short, single-instruction skills, but not enough for the longer-horizon,
+multi-modal robot policies the field is moving toward (high-level planning,
+memory, interjections, VQA, tool use). To support those policies without
+forking the dataset format, LeRobot extends `LeRobotDataset` with two optional
+language columns and a small recipe layer that turns those rows into
+chat-style training samples on the fly.
+
+The design splits cleanly into three layers:
+
+1. **Data in the dataset** — language annotations stored next to frames in
+   `data/chunk-*/file-*.parquet` as two optional columns (`language_persistent`
+   and `language_events`). Datasets without these columns keep their existing
+   behavior.
+2. **Recipe** — a YAML file that declares which annotation rows to bind and
+   how to lay them out as chat turns (`role`, `content`, optional images,
+   optional tool calls). Recipes are pure config; no Python required to add a
+   new one.
+3. **Training format** — at sample time, `RenderMessagesStep` resolves the
+   recipe against the per-frame annotations and emits HF-style `messages` plus
+   LeRobot-specific sidecars (`message_streams`, `target_message_indices`)
+   that policy processors consume.
+
+This page describes each layer in turn.
+
+## Layer 1 — language columns in the dataset
+
+The two optional columns live next to frame data in
+`data/chunk-*/file-*.parquet`:
+
+- `language_persistent`: a list of rows broadcast across every frame in an episode for state that remains active, such as `subtask`, `plan`, and `memory`.
+- `language_events`: a list of rows only on the exact frame where an event was emitted, such as `interjection`, `vqa`, and speech tool calls.
+
+Both columns share the same row shape (event rows omit `timestamp` because the
+frame the row sits on already provides it):
+
+```text
+role: string
+content: string | null
+style: string | null
+timestamp: float32        # persistent rows only
+camera: string | null     # observation.images.* feature key, view-dependent rows only
+tool_calls: list[Json] | null
+```
+
+The `camera` field tags rows whose `content` is grounded in a specific camera
+view. Rows of view-dependent styles (`vqa` and `trace`) MUST set `camera` to
+the matching `observation.images.*` feature key. Rows of every other style —
+including `motion`, which describes robot-frame primitives in joint / Cartesian
+terms — MUST leave `camera` as `null`. Pipeline writers and the validator
+enforce this via `validate_camera_field(style, camera)`.
+
+`meta/tasks.parquet` remains the canonical source for the task. The special `${task}` recipe binding always reads that task string and does not depend on language annotations.
+
+### Architecture
+
+The language stack itself has three internal modules backing layer 1:
+
+1. `lerobot.datasets.language` defines the schema, style registry, and `column_for_style`.
+2. `lerobot.datasets.language_render` resolves rows and renders messages.
+3. `RenderMessagesStep` turns dataset samples into `messages`, `message_streams`, and `target_message_indices`.
+
+`LeRobotDataset` stays recipe-agnostic. It passes `language_persistent` and `language_events` through when present, and unannotated datasets keep their existing behavior.
+
+## Layer 2 — recipe anatomy
+
+Recipes are YAML files backed by `TrainingRecipe` and `MessageTurn`. They
+declare which annotation rows to pull (via `bindings`) and how to compose them
+into chat turns (`messages`).
+
+```yaml
+messages:
+  - { role: user, content: "${task}", stream: high_level }
+  - { role: assistant, content: "${subtask}", stream: low_level, target: true }
+```
+
+A recipe can also branch into a weighted **blend** of sub-recipes. At sample
+time, exactly one branch is selected deterministically from the sample index,
+so different frames train different objectives (e.g. memory updates vs.
+low-level execution vs. VQA) without any Python wiring.
+
+### Temporal semantics
+
+Persistent styles are active after emission until replaced:
+
+- `active_at(t, style=subtask)`
+- `nth_prev(style=memory, offset=1)`
+- `nth_next(style=subtask, offset=1)`
+
+Event styles only exist on their exact timestamp:
+
+- `emitted_at(t, style=interjection)`
+- `emitted_at(t, style=vqa, role=user, camera=observation.images.top)`
+- `emitted_at(t, role=assistant, tool_name=say)`
+
+Exact event matching has no tolerance window, so writers must stamp event rows with frame timestamps from the parquet data.
+
+### View-dependent resolution
+
+For view-dependent styles (`vqa` and `trace`), the resolver gains a
+`camera=` filter parallel to `role=` and `tool_name=`. Datasets with multiple
+cameras typically emit one (`vqa`, `user`) + (`vqa`, `assistant`) pair per
+camera at the same timestamp; without `camera=`, those resolvers see two
+matches and raise an ambiguity error. Recipes consume each camera through its
+own binding plus a matching image block, e.g.
+
+```yaml
+ask_vqa_top:
+  bindings:
+    vqa_query: "emitted_at(t, style=vqa, role=user, camera=observation.images.top)"
+    vqa: "emitted_at(t, style=vqa, role=assistant, camera=observation.images.top)"
+  messages:
+    - role: user
+      stream: high_level
+      if_present: vqa_query
+      content:
+        - { type: image, feature: observation.images.top }
+        - { type: text, text: "${vqa_query}" }
+    - {
+        role: assistant,
+        content: "${vqa}",
+        stream: high_level,
+        target: true,
+        if_present: vqa,
+      }
+```
+
+Add one such sub-recipe per camera the dataset records.
+
+## Layer 3 — training format
+
+Rendered samples use HF-style chat messages plus LeRobot sidecars:
+
+```python
+sample["messages"]
+sample["message_streams"]
+sample["target_message_indices"]
+```
+
+The renderer does not apply a tokenizer chat template. Policy processors decide how to serialize the messages for their backbone, which keeps the same dataset usable across SmolVLA, Pi0.5, and any future VLM that expects OpenAI-style chat messages.
+
+## Graceful absence
+
+If both language columns are missing, `None`, or empty, `RenderMessagesStep` is a no-op.
+If an event-scoped branch is selected on a frame without the required event row, rendering returns `None`, allowing a loader to retry another sample.

docs/source/lerobot-dataset-v3.mdx

@@ -10,6 +10,7 @@ This docs will guide you to:
 - Stream datasets without downloading using `StreamingLeRobotDataset`
 - Apply image transforms for data augmentation during training
 - Migrate existing `v2.1` datasets to `v3.0`
+- Experiment with other `LeRobotDataset` formats and implementations like Lance
 
 ## What’s new in `v3`
 
@@ -43,7 +44,7 @@ lerobot-record \
   --dataset.num_episodes=5 \
   --dataset.single_task="Grab the black cube" \
   --dataset.streaming_encoding=true \
-  # --dataset.vcodec=auto \
+  # --dataset.camera_encoder.vcodec=auto \
   --dataset.encoder_threads=2
 ```
 
@@ -315,3 +316,39 @@ Dataset v3.0 uses incremental parquet writing with buffered metadata for efficie
 - Ensures the dataset is valid for loading
 
 Without calling `finalize()`, your parquet files will be incomplete and the dataset won't load properly.
+
+## Other formats and implementations
+
+### Lance
+
+Lance is a useful format for multimodal AI datasets, especially for large-scale training requiring high performance IO and random access.
+
+The `lerobot-lancedb` package implements `LeRobotLanceDataset` (for JPEG images) and `LeRobotLanceVideoDataset` (for mp4 videos).
+Those two storage layouts both subclass LeRobotDataset and can provide data loading speed ups.
+
+`LeRobotLanceDataset` is a drop-in replacement for `LeRobotDataset`:
+
+```python
+from lerobot.datasets import LeRobotDatasetMetadata
+from lerobot.policies.diffusion.configuration_diffusion import DiffusionConfig
+from lerobot_lancedb import LeRobotLanceDataset, LeRobotLanceVideoDataset
+
+cfg = DiffusionConfig(...)
+meta = LeRobotDatasetMetadata(root=local_dataset_path)  # or use repo_id=... to load metadata from the Hub
+delta_timestamps = {...}
+
+# Use LeRobotLanceDataset for image datasets
+dataset = LeRobotLanceDataset(
+    root=local_dataset_path,                            # or use repo_id=... to stream from the Hub
+    delta_timestamps=delta_timestamps,
+    return_uint8=True,
+)
+# Or use LeRobotLanceVideoDataset for video datasets:
+dataset = LeRobotLanceVideoDataset(
+    root=local_dataset_path,                            # or use repo_id=... to stream from the Hub
+    delta_timestamps=delta_timestamps,
+    return_uint8=True,
+)
+```
+
+Join the discussion on [Github](https://github.com/huggingface/lerobot/issues/3608) and explore the `lerobot-lancedb` documentation [here](https://lancedb.github.io/lerobot-lancedb/).

docs/source/molmoact2.mdx

@@ -0,0 +1,433 @@
+# MolmoAct2 Policy
+
+MolmoAct2 is the LeRobot policy implementation of
+[MolmoAct2](https://allenai.org/blog/molmoact2), ported into the LeRobot
+training, evaluation, checkpointing, and dataset interfaces for easier use with
+LeRobot datasets.
+
+This implementation currently supports training and evaluation for the regular
+MolmoAct2 model. MolmoAct2-Think, which supports adaptive depth reasoning, is
+not included in this LeRobot policy yet and is coming soon.
+
+For the original MolmoAct2 training code used for the experiments reported in
+the paper, see [allenai/molmoact2](https://github.com/allenai/molmoact2).
+
+## Installation Requirements
+
+Install LeRobot with the MolmoAct2 optional dependencies:
+
+```bash
+pip install -e ".[molmoact2]"
+```
+
+To run the models in this repository, you need an NVIDIA GPU. The measurements
+below were taken on a single NVIDIA H100 80GB with bf16 model loading, LIBERO with two RGB cameras. MolmoAct2 rows use `chunk_size=10`, action dim 7
+padded to `expected_max_action_dim=32`, and `num_flow_timesteps=8`. Training measurements use
+`gradient_checkpointing=true` and include the forward pass, backward pass,
+gradient clipping, optimizer step, and optimizer state allocation. Values are
+peak GPU memory sampled with `nvidia-smi`. Leave a few GiB of headroom for
+dataloader workers, CUDA context, and fragmentation.
+
+Multi-GPU training through `accelerate` increases throughput and global batch
+size, but this LeRobot port does not currently expose the original MolmoAct2
+`fsdp_devices` model-parallel training path. The current training script has
+not been tested for multi-node training.
+
+| Mode                                             | Peak Memory, bs=8 | Peak Memory, bs=16 | Peak Memory, bs=32 |
+| ------------------------------------------------ | ----------------: | -----------------: | -----------------: |
+| Inference, continuous, CUDA graph enabled (bs=1) |          12.1 GiB |                  - |                  - |
+| Fine-tuning, action expert only, continuous      |          16.5 GiB |           18.3 GiB |           21.4 GiB |
+| Fine-tuning, LoRA VLM, both action modes         |          20.2 GiB |           26.8 GiB |           41.3 GiB |
+| Fine-tuning, full model, both action modes       |          48.3 GiB |           49.8 GiB |           60.1 GiB |
+
+The repo has been tested with Ubuntu 22.04.
+
+## Usage
+
+To use MolmoAct2 in a LeRobot training config, set:
+
+```python
+policy.type=molmoact2
+```
+
+## Training
+
+MolmoAct2 can be fine-tuned from either the released MolmoAct2 Hugging Face
+checkpoint format or from a checkpoint already saved by LeRobot. Both routes use
+the same LeRobot training loop, dataset transforms, checkpoint saving, and
+logging. The difference is only how the initial policy weights and processor
+state are loaded.
+
+### Training With Original MolmoAct2 Weight
+
+Use `policy.checkpoint_path` when starting from a released MolmoAct2 checkpoint,
+for example `allenai/MolmoAct2` or `allenai/MolmoAct2-LIBERO`. LeRobot will load
+the original HF model files, then build its own policy processor from the
+dataset metadata and the policy options below.
+
+The command below shows full fine-tuning on the merged LIBERO dataset. It uses
+bf16 model loading, 8 flow timesteps, LeRobot dataset statistics, image
+augmentation, and LeRobot's checkpointing/logging path.
+
+```bash
+accelerate launch \
+  --num_processes=8 \
+  --mixed_precision=bf16 \
+  -m lerobot.scripts.lerobot_train \
+  --dataset.repo_id=allenai/MolmoAct2-LIBERO-Dataset \
+  --dataset.root=/path/to/lerobot/data/allenai/MolmoAct2-LIBERO-Dataset \
+  --dataset.video_backend=pyav \
+  --dataset.image_transforms.enable=true \
+  --policy.type=molmoact2 \
+  --policy.checkpoint_path=allenai/MolmoAct2-LIBERO \
+  --policy.device=cuda \
+  --policy.action_mode=both \
+  --policy.chunk_size=10 \
+  --policy.n_action_steps=10 \
+  --policy.setup_type="single franka robotic arm in libero" \
+  --policy.control_mode="delta end-effector pose" \
+  --policy.image_keys='["observation.images.image","observation.images.wrist_image"]' \
+  --policy.model_dtype=bfloat16 \
+  --policy.num_flow_timesteps=8 \
+  --policy.gradient_checkpointing=true \
+  --policy.freeze_embedding=true \
+  --policy.normalize_gripper=false \
+  --policy.enable_knowledge_insulation=false \
+  --policy.push_to_hub=false \
+  --wandb.enable=true \
+  --wandb.entity=<wandb_entity> \
+  --wandb.project=<wandb_project> \
+  --job_name=<job_name> \
+  --output_dir=outputs/<job_name> \
+  --steps=10000 \
+  --batch_size=32 \
+  --num_workers=4 \
+  --log_freq=20 \
+  --eval_freq=-1 \
+  --save_checkpoint=true \
+  --save_freq=2000
+```
+
+### Training With LeRobot MolmoAct2 Weight
+
+Use `policy.path` when starting from a MolmoAct2 checkpoint that was saved by
+LeRobot, either from a local `pretrained_model` directory or from the Hub. This
+restores the saved LeRobot policy config, model weights, processor, and
+normalization statistics. You can still override training-time options such as
+`batch_size`, `steps`, LoRA flags, or `policy.action_mode`.
+
+```bash
+accelerate launch \
+  --num_processes=8 \
+  --mixed_precision=bf16 \
+  -m lerobot.scripts.lerobot_train \
+  --dataset.repo_id=allenai/MolmoAct2-LIBERO-Dataset \
+  --dataset.root=/path/to/lerobot/data/allenai/MolmoAct2-LIBERO-Dataset \
+  --dataset.video_backend=pyav \
+  --dataset.image_transforms.enable=true \
+  --policy.path=/path/to/pretrained_model \
+  --policy.device=cuda \
+  --policy.action_mode=both \
+  --policy.chunk_size=10 \
+  --policy.n_action_steps=10 \
+  --policy.model_dtype=bfloat16 \
+  --policy.num_flow_timesteps=8 \
+  --policy.gradient_checkpointing=true \
+  --wandb.enable=true \
+  --wandb.entity=<wandb_entity> \
+  --wandb.project=<wandb_project> \
+  --job_name=<job_name> \
+  --output_dir=outputs/<job_name> \
+  --steps=10000 \
+  --batch_size=32 \
+  --num_workers=4 \
+  --log_freq=20 \
+  --eval_freq=-1 \
+  --save_checkpoint=true \
+  --save_freq=2000
+```
+
+### Common Practices
+
+For fine-tuning on a comparatively small dataset, such as a single LIBERO suite
+or a real-world dataset with less than 200 demonstrations, a global batch size of
+16 to 32 is a good starting point. In these settings, `policy.enable_lora_vlm=true` or `policy.train_action_expert_only=true` is also a practical choice. In both
+cases, we intentionally keep the action expert fully trainable, which we found
+to be crucial for model performance. For larger fine-tuning datasets, larger
+global batch sizes and full fine-tuning are usually preferred.
+
+### Common Policy Options
+
+- `policy.checkpoint_path`: original MolmoAct2 HF checkpoint to initialize from.
+  Use this for released MolmoAct2 weights.
+- `policy.path`: LeRobot checkpoint to initialize from. Use this for checkpoints
+  created by LeRobot training.
+- `policy.action_mode`: training target, one of `continuous`, `discrete`, or
+  `both`. `both` trains the flow-matching action expert and the discrete
+  action-token loss.
+- `policy.train_action_expert_only`: trains only parameters whose names contain
+  `action_expert`. It requires `policy.action_mode=continuous`.
+- `policy.enable_lora_vlm`: enables LoRA on VLM linear layers. Use
+  `policy.enable_lora_action_expert=true` only if LoRA should also cover action
+  expert linear layers. When `policy.enable_lora_action_expert=false`, the
+  action expert base weights remain fully trainable while the VLM is trained
+  through LoRA adapters. When `policy.enable_lora_action_expert=true`, the
+  action expert is also adapter-tuned instead of fully fine-tuned.
+- `policy.enable_knowledge_insulation`: when `true`, detaches action-expert
+  context K/V states before the action loss. The default is `false`.
+- `policy.chunk_size`: action horizon used by the policy. For LIBERO we use
+  `10`. This LeRobot port overrides the loaded checkpoint's
+  `max_action_horizon` with this value.
+- `policy.n_action_steps`: number of actions consumed from each predicted
+  chunk before querying the policy again. For LIBERO, set it to `chunk_size`.
+- `policy.setup_type`: text inserted into the prompt to describe the robot and
+  scene, e.g. `single franka robotic arm in libero`. More examples are listed
+  in the `metadata_by_tag` entries of
+  [`norm_stats.json`](https://huggingface.co/allenai/MolmoAct2/blob/main/norm_stats.json).
+- `policy.control_mode`: text inserted into the prompt to describe the action
+  space, e.g. `delta end-effector pose` or `absolute joint pose`.
+- `policy.image_keys`: ordered LeRobot image observation keys passed to the
+  processor.
+- `policy.model_dtype`: checkpoint/forward dtype, one of `float32`,
+  `bfloat16`, or `float16`. Use `bfloat16` for normal training.
+- `policy.num_flow_timesteps`: number of flow-matching timesteps sampled per
+  example during training. We use `8` for fine-tuning.
+- `policy.num_inference_steps`: optional override for continuous action
+  generation steps at inference time.
+- `policy.gradient_checkpointing`: enables checkpointing in the VLM/action path
+  to reduce activation memory.
+- `policy.freeze_embedding`: freezes input embeddings. The default is `true`.
+- `policy.normalize_gripper`: controls whether gripper dimensions are included
+  in state/action quantile normalization. The default is `false`.
+- `policy.normalize_language`: normalizes task strings before prompt
+  construction. The default is `true`.
+- `policy.mask_action_dim_padding`: masks padded dimensions in the flow loss.
+  Released checkpoints use `policy.expected_max_action_dim=32`.
+- `policy.max_sequence_length`: optional manual sequence cap. Leave unset to
+  infer it from images, state dimension, action dimension, action horizon, and
+  discrete-action mode.
+
+### Learning Rates
+
+MolmoAct2 uses parameter-group learning rates to match the original MolmoAct2
+fine-tuning experiments.
+
+- Full fine-tuning uses `policy.optimizer_lr=1e-5` for the VLM,
+  `policy.optimizer_vit_lr=5e-6` for the vision tower,
+  `policy.optimizer_connector_lr=5e-6` for image connector layers, and
+  `policy.optimizer_action_expert_lr=5e-5` for the action expert.
+- LoRA VLM fine-tuning sets the VLM, vision, and connector LoRA parameter
+  groups to `5e-5` when `policy.enable_lora_vlm=true`. By default,
+  `policy.enable_lora_action_expert=false`, so the action expert is still fully
+  fine-tuned with `policy.optimizer_action_expert_lr`. If
+  `policy.enable_lora_action_expert=true`, the action expert is trained through
+  LoRA adapters instead.
+- Action-expert-only fine-tuning trains only the action expert and uses
+  `policy.optimizer_action_expert_lr=5e-5`.
+
+You can override the full fine-tuning and action-expert learning rates with
+`policy.optimizer_lr`, `policy.optimizer_vit_lr`,
+`policy.optimizer_connector_lr`, and `policy.optimizer_action_expert_lr`.
+Scheduler settings can be changed with `policy.scheduler_warmup_steps`,
+`policy.scheduler_decay_steps`, and `policy.scheduler_decay_lr`.
+
+### Dataset Quantile Statistics
+
+MolmoAct2 defaults to quantile normalization for state and action features. If
+your dataset has not been converted with quantile statistics, you can add them
+with:
+
+```bash
+python src/lerobot/datasets/v30/augment_dataset_quantile_stats.py \
+  --repo-id=your_dataset
+```
+
+Alternatively, train MolmoAct2 with mean/std normalization:
+
+```bash
+--policy.normalization_mapping='{"ACTION": "MEAN_STD", "STATE": "MEAN_STD", "VISUAL": "IDENTITY"}'
+```
+
+## Evaluation
+
+Evaluation also supports both LeRobot-saved checkpoints and original MolmoAct2
+HF checkpoints. For LIBERO replication, keep the EGL rendering environment
+fixed and use `policy.per_episode_seed=true`.
+
+**Important:** We found that `num_steps_wait=10` does not reliably let the
+LIBERO scene stabilize and can degrade measured success. All LIBERO evaluation
+results reported here use `num_steps_wait=50`.
+
+### Evaluation With LeRobot MolmoAct2 Weight
+
+Use `policy.path` for a checkpoint saved by LeRobot. The saved processor and
+normalization statistics are restored together with the model.
+
+```bash
+export MUJOCO_GL=egl
+export PYOPENGL_PLATFORM=egl
+export OMP_NUM_THREADS=1
+export MKL_NUM_THREADS=1
+
+lerobot-eval \
+  --policy.path=allenai/MolmoAct2-LIBERO-LeRobot \
+  --policy.inference_action_mode=continuous \
+  --policy.model_dtype=bfloat16 \
+  --policy.use_amp=true \
+  --policy.enable_inference_cuda_graph=true \
+  --policy.device=cuda \
+  --policy.per_episode_seed=true \
+  --policy.eval_seed=1000 \
+  --env.type=libero \
+  --env.task=libero_10,libero_goal,libero_object,libero_spatial \
+  --env.camera_name_mapping='{"agentview_image":"image","robot0_eye_in_hand_image":"wrist_image"}' \
+  --eval.batch_size=1 \
+  --eval.n_episodes=50 \
+  --seed=1000
+```
+
+### Evaluation With Original MolmoAct2 Weight
+
+You can evaluate a released Hugging Face checkpoint directly without first
+converting it to a LeRobot checkpoint. In this case, set
+`policy.checkpoint_path` to the HF model repo and provide `policy.norm_tag`.
+For LIBERO, `policy.norm_tag=libero` loads the LIBERO action/state
+normalization statistics, action horizon, prompt metadata, and image-key order
+from the checkpoint's `norm_stats.json`.
+
+To fully replicate the MolmoAct2 paper results with released Hugging Face
+checkpoints, we recommend using the v0.5.1-pinned
+[`allenai/lerobot` `molmoact2-hf-inference`](https://github.com/allenai/lerobot/tree/molmoact2-hf-inference)
+branch. That branch matches the original evaluation settings used for the
+reported numbers.
+
+```bash
+export MUJOCO_GL=egl
+export PYOPENGL_PLATFORM=egl
+export OMP_NUM_THREADS=1
+export MKL_NUM_THREADS=1
+
+lerobot-eval \
+  --policy.type=molmoact2 \
+  --policy.checkpoint_path=allenai/MolmoAct2-LIBERO \
+  --policy.norm_tag=libero \
+  --policy.inference_action_mode=continuous \
+  --policy.model_dtype=float32 \
+  --policy.use_amp=false \
+  --policy.enable_inference_cuda_graph=true \
+  --policy.device=cuda \
+  --policy.per_episode_seed=true \
+  --policy.eval_seed=1000 \
+  --env.type=libero \
+  --env.task=libero_goal \
+  --env.camera_name_mapping='{"agentview_image":"image","robot0_eye_in_hand_image":"wrist_image"}' \
+  --eval.batch_size=1 \
+  --eval.n_episodes=50 \
+  --seed=1000
+```
+
+Use `--env.task=libero_10,libero_goal,libero_object,libero_spatial` to run the
+full LIBERO suite. The same command works for other released MolmoAct2
+checkpoints as long as the requested `policy.norm_tag` exists in that
+checkpoint's `norm_stats.json`.
+
+### Common Evaluation Options
+
+- `policy.inference_action_mode`: required for rollout. Use `continuous` for
+  flow-matching inference or `discrete` for action-token inference. It must be
+  compatible with the training-time `policy.action_mode` saved in the
+  checkpoint.
+- `policy.path`: LeRobot checkpoint path or Hub repo. Use this for checkpoints
+  saved by LeRobot.
+- `policy.checkpoint_path`: original MolmoAct2 HF checkpoint path or Hub repo.
+  Use this with `policy.type=molmoact2` and `policy.norm_tag`.
+- `policy.norm_tag`: selects normalization statistics, prompt metadata,
+  image-key order, and action horizon from the original checkpoint's
+  `norm_stats.json`. It is required for direct original-HF checkpoint
+  evaluation.
+- `policy.model_dtype`: model load/forward dtype. Use `bfloat16` for normal
+  GPU evaluation. Use `float32` only when you explicitly want fp32 inference.
+- `policy.use_amp`: runs the policy forward under autocast during eval. For
+  `model_dtype=bfloat16`, keep this enabled.
+- `policy.enable_inference_cuda_graph`: enables the MolmoAct2 inference CUDA
+  graph path for faster repeated continuous-action rollout.
+- `policy.per_episode_seed` and `policy.eval_seed`: make stochastic continuous
+  action generation deterministic per episode for replication.
+- `env.task`: comma-separated LIBERO suites or a single suite. Use
+  `libero_10,libero_goal,libero_object,libero_spatial` for the full benchmark.
+- `env.camera_name_mapping`: maps LIBERO camera names to the image keys expected
+  by the policy processor.
+
+## Performance Results
+
+### LIBERO Benchmark Results
+
+MolmoAct2 has demonstrated strong performance on the LIBERO benchmark suite. To
+compare and test its LeRobot implementation, we fine-tuned
+[`allenai/MolmoAct2-LIBERO`](https://huggingface.co/allenai/MolmoAct2-LIBERO)
+for an additional 10k steps on the LIBERO dataset with per-GPU batch size 32 on
+8 H100 GPUs, then compared the results to the original MolmoAct2 reference
+results.
+
+The LeRobot fine-tuned checkpoint reported here is available at
+[`allenai/MolmoAct2-LIBERO-LeRobot`](https://huggingface.co/allenai/MolmoAct2-LIBERO-LeRobot)
+and was trained on
+[`allenai/MolmoAct2-LIBERO-Dataset`](https://huggingface.co/datasets/allenai/MolmoAct2-LIBERO-Dataset).
+
+| Benchmark      | LeRobot Implementation | MolmoAct2 Original |
+| -------------- | ---------------------: | -----------------: |
+| LIBERO Spatial |                  98.4% |              97.8% |
+| LIBERO Object  |                 100.0% |             100.0% |
+| LIBERO Goal    |                  98.0% |              97.8% |
+| LIBERO 10      |                  96.6% |              93.2% |
+| Average        |                 98.25% |             97.20% |
+
+These results demonstrate MolmoAct2's strong performance across diverse robotic
+manipulation tasks. To reproduce them, follow the instructions in the LIBERO
+evaluation section.
+
+## Differences From the Original Implementation
+
+This LeRobot port is intended to match MolmoAct2 behavior while using LeRobot's
+dataset, training, evaluation, checkpoint, and logging infrastructure. The main
+differences from the original training repository are:
+
+- The original paper training stack loads the model in fp32 and trains under
+  mixed precision. This LeRobot port usually loads the checkpoint directly in
+  `policy.model_dtype=bfloat16` for lower memory use.
+- The original repository uses its own FSDP/model-parallel training path. The
+  LeRobot port uses the standard LeRobot/Accelerate training path and has not
+  been tested for multi-node training.
+- The original repository supports sequence packing. The LeRobot port trains on
+  one LeRobot sample per item and pads to an inferred fixed sequence budget.
+- The LeRobot port follows LeRobot's optimizer, scheduler, checkpoint saving,
+  dataset transforms, image augmentation, and Weights & Biases logging
+  conventions.
+- The original training path supports mixed action horizons by padding to
+  `max_action_horizon` and masking padded horizon slots in the action expert
+  self-attention. This is useful when training across datasets with different
+  control frequencies. The LeRobot port currently targets single-dataset
+  fine-tuning, so `policy.chunk_size` overrides the checkpoint
+  `max_action_horizon` and horizon masking is not implemented yet. Support for
+  this mixed-horizon path is planned.
+
+## Citation
+
+```bibtex
+@misc{fang2026molmoact2actionreasoningmodels,
+      title={MolmoAct2: Action Reasoning Models for Real-world Deployment},
+      author={Haoquan Fang and Jiafei Duan and Donovan Clay and Sam Wang and Shuo Liu and Weikai Huang and Xiang Fan and Wei-Chuan Tsai and Shirui Chen and Yi Ru Wang and Shanli Xing and Jaemin Cho and Jae Sung Park and Ainaz Eftekhar and Peter Sushko and Karen Farley and Angad Wadhwa and Cole Harrison and Winson Han and Ying-Chun Lee and Eli VanderBilt and Rose Hendrix and Suveen Ellawela and Lucas Ngoo and Joyce Chai and Zhongzheng Ren and Ali Farhadi and Dieter Fox and Ranjay Krishna},
+      year={2026},
+      eprint={2605.02881},
+      archivePrefix={arXiv},
+      primaryClass={cs.RO},
+      url={https://arxiv.org/abs/2605.02881},
+}
+```
+
+## License
+
+This model is licensed under Apache 2.0. It is intended for research and
+educational use in accordance with
+[Ai2's Responsible Use Guidelines](https://allenai.org/responsible-use),
+consistent with [allenai/molmoact2](https://github.com/allenai/molmoact2).

docs/source/peft_training.mdx

@@ -28,13 +28,15 @@ lerobot-train \
  --steps=100000 \
  --batch_size=32 \
  --peft.method_type=LORA \
- --peft.r=64
+ --peft.r=64 \
+ --peft.lora_alpha=64
 ```
 
 Note the `--peft.method_type` parameter that let's you select which PEFT method to use. Here we use
 [LoRA](https://huggingface.co/docs/peft/main/en/package_reference/lora) (Low-Rank Adapter) which is probably the most
 popular fine-tuning method to date. Low-rank adaption means that we only fine-tune a matrix with comparably low rank
-instead of the full weight matrix. This rank can be specified using the `--peft.r` parameter. The higher the rank
+instead of the full weight matrix. This rank can be specified using the `--peft.r` parameter, and the LoRA scaling factor with
+`--peft.lora_alpha` (where `scaling = lora_alpha / r`). The higher the rank
 the closer you get to full fine-tuning
 
 There are more complex methods that have more parameters. These are not yet supported, feel free to raise an issue

docs/source/policy_molmoact2_README.md

@@ -0,0 +1,39 @@
+# MolmoAct2
+
+This repository contains the LeRobot policy implementation of
+[MolmoAct2](https://allenai.org/blog/molmoact2), ported into LeRobot for
+training, evaluation, checkpointing, and dataset compatibility.
+
+This implementation currently supports training and evaluation for the regular
+MolmoAct2 model. MolmoAct2-Think, which supports adaptive depth reasoning, is
+not included in this LeRobot policy yet and is coming soon.
+
+For the original MolmoAct2 training code used for the experiments reported in
+the paper, see [allenai/molmoact2](https://github.com/allenai/molmoact2).
+
+## LIBERO Evaluation
+
+Important: we found that `num_steps_wait=10` does not reliably let the LIBERO
+scene stabilize and can degrade measured success. All LIBERO evaluation results
+reported for this LeRobot implementation use `num_steps_wait=50`.
+
+## Citation
+
+```bibtex
+@misc{fang2026molmoact2actionreasoningmodels,
+      title={MolmoAct2: Action Reasoning Models for Real-world Deployment},
+      author={Haoquan Fang and Jiafei Duan and Donovan Clay and Sam Wang and Shuo Liu and Weikai Huang and Xiang Fan and Wei-Chuan Tsai and Shirui Chen and Yi Ru Wang and Shanli Xing and Jaemin Cho and Jae Sung Park and Ainaz Eftekhar and Peter Sushko and Karen Farley and Angad Wadhwa and Cole Harrison and Winson Han and Ying-Chun Lee and Eli VanderBilt and Rose Hendrix and Suveen Ellawela and Lucas Ngoo and Joyce Chai and Zhongzheng Ren and Ali Farhadi and Dieter Fox and Ranjay Krishna},
+      year={2026},
+      eprint={2605.02881},
+      archivePrefix={arXiv},
+      primaryClass={cs.RO},
+      url={https://arxiv.org/abs/2605.02881},
+}
+```
+
+## License
+
+This model is licensed under Apache 2.0. It is intended for research and
+educational use in accordance with
+[Ai2's Responsible Use Guidelines](https://allenai.org/responsible-use),
+consistent with [allenai/molmoact2](https://github.com/allenai/molmoact2).

docs/source/reachy2.mdx

@@ -161,7 +161,7 @@ lerobot-record \
     --dataset.private=true \
     --dataset.streaming_encoding=true \
     --dataset.encoder_threads=2 \
-    # --dataset.vcodec=auto \
+    # --dataset.camera_encoder.vcodec=auto \
     --display_data=true
 ```
 
@@ -203,7 +203,7 @@ lerobot-record \
     --dataset.private=true \
     --dataset.streaming_encoding=true \
     --dataset.encoder_threads=2 \
-    # --dataset.vcodec=auto \
+    # --dataset.camera_encoder.vcodec=auto \
     --display_data=true
 ```
 

docs/source/rebot_b601.mdx

@@ -0,0 +1,186 @@
+# reBot B601-DM
+
+[reBot B601-DM](https://wiki.seeedstudio.com/rebot_arm_b601_dm_lerobot/) is an open-source, low-cost robot arm from Seeed Studio for embodied-AI and imitation learning. It comes as a **follower** arm (the `B601-DM`, a 6-DOF arm plus gripper driven by Damiao CAN motors) and a **leader** arm (the `StarArm102` / `reBot Arm 102`, driven by FashionStar UART smart servos) used to teleoperate it.
+
+This page covers **calibration** and **teleoperation** for both single-arm and bimanual (dual-arm) setups.
+
+<div style="display: flex; align-items: center; gap: 10px;">
+  <img
+    src="https://files.seeedstudio.com/wiki/robotics/projects/lerobot/b601dm_zeroposition.jpg"
+    alt="reBot B601-DM follower arm at its zero position"
+    width="48%"
+  />
+  <img
+    src="https://files.seeedstudio.com/wiki/robotics/projects/lerobot/102_zeroposition.jpg"
+    alt="reBot Arm 102 leader arm at its zero position"
+    width="48%"
+  />
+</div>
+
+_Left: the B601-DM follower at its zero position. Right: the reBot Arm 102 leader at its zero position. Images courtesy of [Seeed Studio](https://wiki.seeedstudio.com/rebot_arm_b601_dm_lerobot/)._
+
+## Install LeRobot 🤗
+
+Follow our [Installation Guide](./installation), then install the reBot support:
+
+```bash
+pip install -e ".[rebot]"
+```
+
+This pulls in `motorbridge` (CAN motor control for the B601-DM follower) and `motorbridge-smart-servo` (FashionStar UART servos for the reBot Arm 102 leader).
+
+## Registered device types
+
+| Type                     | Kind                                         |
+| ------------------------ | -------------------------------------------- |
+| `rebot_b601_follower`    | single-arm B601-DM follower robot            |
+| `bi_rebot_b601_follower` | bimanual (dual-arm) follower robot           |
+| `rebot_102_leader`       | single-arm reBot Arm 102 leader teleoperator |
+| `bi_rebot_102_leader`    | bimanual (dual-arm) leader teleoperator      |
+
+The bimanual types compose two single-arm instances and namespace each arm's
+observation/action keys with a `left_` / `right_` prefix. Per-arm settings are
+passed through nested `left_arm_config.*` / `right_arm_config.*` arguments.
+
+## Find the USB ports
+
+For each device, find the USB port associated with its motor bus using:
+
+```bash
+lerobot-find-port
+```
+
+<Tip warning={true}>
+  On Linux, remove `brltty` (`sudo apt remove brltty`) so it does not hold the
+  leader's USB serial port. You may also need to grant access to the serial
+  devices: `sudo chmod 666 /dev/ttyACM* /dev/ttyUSB*`.
+</Tip>
+
+## Calibration
+
+Neither arm stores a persistent hardware calibration: every time it connects, the motors are re-zeroed against the pose the arm is physically holding. Calibration simply records that zero pose. When prompted, **manually move the arm to its zero position** (the default sit-down pose shown above, gripper fully closed) and press <kbd>ENTER</kbd>.
+
+### Follower (B601-DM)
+
+<hfoptions id="calibrate-follower">
+<hfoption id="Single arm">
+
+```bash
+lerobot-calibrate \
+    --robot.type=rebot_b601_follower \
+    --robot.port=/dev/ttyACM0 \
+    --robot.id=follower \
+    --robot.can_adapter=damiao
+```
+
+</hfoption>
+<hfoption id="Dual arm">
+
+Connect the bimanual follower; calibration runs for the left arm, then the right arm.
+
+```bash
+lerobot-calibrate \
+    --robot.type=bi_rebot_b601_follower \
+    --robot.id=bi_follower \
+    --robot.left_arm_config.port=/dev/ttyACM0 \
+    --robot.left_arm_config.can_adapter=damiao \
+    --robot.right_arm_config.port=/dev/ttyACM1 \
+    --robot.right_arm_config.can_adapter=damiao
+```
+
+Per-arm calibration files are saved with `_left` / `_right` suffixes on the id.
+
+</hfoption>
+</hfoptions>
+
+### Leader (reBot Arm 102)
+
+<hfoptions id="calibrate-leader">
+<hfoption id="Single arm">
+
+```bash
+lerobot-calibrate \
+    --teleop.type=rebot_102_leader \
+    --teleop.port=/dev/ttyUSB0 \
+    --teleop.id=leader
+```
+
+</hfoption>
+<hfoption id="Dual arm">
+
+```bash
+lerobot-calibrate \
+    --teleop.type=bi_rebot_102_leader \
+    --teleop.id=bi_leader \
+    --teleop.left_arm_config.port=/dev/ttyUSB0 \
+    --teleop.right_arm_config.port=/dev/ttyUSB1
+```
+
+</hfoption>
+</hfoptions>
+
+## Teleoperation
+
+Once both arms are calibrated, drive the follower with the leader. The follower talks to its CAN bus through a Damiao serial bridge (`can_adapter=damiao`, the default) or a SocketCAN adapter (`can_adapter=socketcan`). See the [OpenArm page](./openarm) for more details on the SocketCAN adapter configuration.
+
+<hfoptions id="teleoperate">
+<hfoption id="Single arm">
+
+```bash
+lerobot-teleoperate \
+    --robot.type=rebot_b601_follower \
+    --robot.port=/dev/ttyACM0 \
+    --robot.id=follower \
+    --robot.can_adapter=damiao \
+    --teleop.type=rebot_102_leader \
+    --teleop.port=/dev/ttyUSB0 \
+    --teleop.id=leader
+```
+
+</hfoption>
+<hfoption id="Dual arm">
+
+The bimanual leader and follower reuse the single-arm classes; each arm is
+configured through nested `left_arm_config.*` / `right_arm_config.*` arguments,
+so a bimanual reBot Arm 102 leader drives a bimanual B601-DM follower.
+
+```bash
+lerobot-teleoperate \
+    --robot.type=bi_rebot_b601_follower \
+    --robot.id=bi_follower \
+    --robot.left_arm_config.port=/dev/ttyACM0 \
+    --robot.left_arm_config.can_adapter=damiao \
+    --robot.right_arm_config.port=/dev/ttyACM1 \
+    --robot.right_arm_config.can_adapter=damiao \
+    --teleop.type=bi_rebot_102_leader \
+    --teleop.id=bi_leader \
+    --teleop.left_arm_config.port=/dev/ttyUSB0 \
+    --teleop.right_arm_config.port=/dev/ttyUSB1
+```
+
+</hfoption>
+</hfoptions>
+
+<Tip>
+  The leader and follower share the same joint names (`shoulder_pan,
+  shoulder_lift, elbow_flex, wrist_flex, wrist_yaw, wrist_roll, gripper`), so
+  leader actions map directly onto the follower.
+</Tip>
+
+If the motion of a joint is reversed, flip its sign in the leader's `joint_directions` (the gripper also carries a scale to widen its range to the follower):
+
+```bash
+lerobot-teleoperate \
+    --robot.type=rebot_b601_follower \
+    --robot.port=/dev/ttyACM0 \
+    --robot.can_adapter=damiao \
+    --teleop.type=rebot_102_leader \
+    --teleop.port=/dev/ttyUSB0 \
+    --teleop.joint_directions='{"shoulder_pan":-1,"shoulder_lift":-1,"elbow_flex":1,"wrist_flex":1,"wrist_yaw":1,"wrist_roll":-1,"gripper":-6}'
+```
+
+## Recording datasets
+
+Swap `lerobot-teleoperate` for `lerobot-record` (with the same `--robot.*` / `--teleop.*` arguments, plus `--dataset.*`) to record demonstrations for training. See [Imitation Learning for Robots](./il_robots) for the full workflow.
+
+For hardware assembly and wiring, see the [Seeed Studio reBot wiki](https://wiki.seeedstudio.com/rebot_arm_b601_dm_lerobot/).

docs/source/robometer.mdx

@@ -0,0 +1,185 @@
+# ROBOMETER
+
+ROBOMETER is a **general-purpose video-language robotic reward model**. It predicts dense, frame-level task progress and frame-level success from a trajectory video and a task description.
+
+**Paper**: [ROBOMETER: Scaling General-Purpose Robotic Reward Models via Trajectory Comparisons](https://arxiv.org/abs/2603.02115)
+**Project**: [robometer.github.io](https://robometer.github.io/)
+**Original code**: [github.com/robometer/robometer](https://github.com/robometer/robometer)
+**Checkpoint**: [lerobot/Robometer-4B](https://huggingface.co/lerobot/Robometer-4B)
+
+## Overview
+
+ROBOMETER builds on `Qwen/Qwen3-VL-4B-Instruct` and adds three lightweight prediction heads:
+
+- **Progress head**: predicts per-frame task progress in `[0, 1]`.
+- **Success head**: predicts per-frame task success probability.
+- **Preference head**: predicts which of two trajectories better completes the task during training.
+
+The paper trains ROBOMETER with a composite objective:
+
+```text
+L = L_pref + L_prog + L_succ
+```
+
+The LeRobot integration is currently **inference-only**. It preserves the preference head so that the published `Robometer-4B` checkpoint loads without remapping, but `compute_reward()` queries the progress or success head only.
+
+## What the LeRobot Integration Covers
+
+- Standard `reward_model.type=robometer` configuration through LeRobot.
+- Qwen3-VL image and text preprocessing through `RobometerEncoderProcessorStep`.
+- LeRobot reward-model save/load APIs through `PreTrainedRewardModel`.
+- Dense, frame-level progress and success predictions internally.
+- A scalar reward through `compute_reward()` for downstream LeRobot reward-model usage.
+
+This page focuses on using the published ROBOMETER checkpoint as a zero-shot reward model. Training ROBOMETER from scratch is outside the current LeRobot integration.
+
+## Installation Requirements
+
+1. Install LeRobot by following the [Installation Guide](./installation).
+2. Install the ROBOMETER dependencies:
+
+```bash
+pip install -e ".[robometer]"
+```
+
+If you use `uv` directly from a source checkout:
+
+```bash
+uv sync --extra robometer
+```
+
+ROBOMETER uses a Qwen3-VL-4B backbone, so GPU inference is strongly recommended.
+
+## Model Inputs and Outputs
+
+ROBOMETER expects:
+
+- A trajectory video or sequence of frames.
+- A natural-language task description.
+
+In LeRobot datasets, the preprocessor reads:
+
+| Config field              | Default                  | Meaning                                               |
+| ------------------------- | ------------------------ | ----------------------------------------------------- |
+| `reward_model.image_key`  | `observation.images.top` | Camera/video observation used by ROBOMETER            |
+| `reward_model.task_key`   | `task`                   | Key in complementary data that stores the task string |
+| `reward_model.max_frames` | `8`                      | Maximum number of frames passed to ROBOMETER          |
+
+The model predicts per-frame progress and success internally. The LeRobot reward API returns a scalar per sample:
+
+- `reward_output="progress"` (default): return the last-frame progress, clamped to `[0, 1]`.
+- `reward_output="success"`: return `1.0` if the last-frame success probability is above `success_threshold`, otherwise `0.0`.
+
+## Usage
+
+### Load the Reward Model Directly
+
+```python
+from lerobot.rewards.robometer import RobometerConfig, RobometerRewardModel
+
+cfg = RobometerConfig(
+    pretrained_path="lerobot/Robometer-4B",
+    device="cuda",
+    reward_output="progress",
+)
+reward_model = RobometerRewardModel.from_pretrained(cfg.pretrained_path, config=cfg)
+```
+
+### Encode Frames and Compute a Reward
+
+For a direct Python call, provide frames as `uint8` arrays with shape `(T, H, W, C)` and a task string:
+
+```python
+from lerobot.rewards.robometer.modeling_robometer import ROBOMETER_FEATURE_PREFIX
+from lerobot.rewards.robometer.processor_robometer import RobometerEncoderProcessorStep
+
+# frames: np.ndarray, shape (T, H, W, C), dtype uint8
+# task: str
+encoder = RobometerEncoderProcessorStep(
+    base_model_id=cfg.base_model_id,
+    use_multi_image=cfg.use_multi_image,
+    use_per_frame_progress_token=cfg.use_per_frame_progress_token,
+    max_frames=cfg.max_frames,
+)
+
+encoded = encoder.encode_samples([(frames, task)])
+batch = {f"{ROBOMETER_FEATURE_PREFIX}{key}": value for key, value in encoded.items()}
+
+reward = reward_model.compute_reward(batch)
+```
+
+`reward` is a tensor of shape `(batch_size,)`.
+
+### Use the Reward Factory
+
+You can also instantiate ROBOMETER through the reward factory:
+
+```python
+from lerobot.rewards import make_reward_model, make_reward_model_config, make_reward_pre_post_processors
+
+cfg = make_reward_model_config(
+    "robometer",
+    pretrained_path="lerobot/Robometer-4B",
+    device="cuda",
+    image_key="observation.images.top",
+)
+reward_model = make_reward_model(cfg)
+preprocessor, postprocessor = make_reward_pre_post_processors(cfg)
+```
+
+The preprocessor writes Qwen-VL tensors under the `observation.robometer.*` namespace, and `compute_reward()` reads those encoded tensors.
+
+## Configuration Notes
+
+### Backbone and Vocabulary
+
+The published checkpoint uses a Qwen3-VL-4B backbone. ROBOMETER adds five special tokens to the tokenizer in a fixed order:
+
+```text
+<|split_token|>
+<|reward_token|>
+<|pref_token|>
+<|sim_token|>
+<|prog_token|>
+```
+
+`<|prog_token|>` is inserted after each frame and is the hidden-state position used for per-frame progress and success prediction. `<|split_token|>` and `<|pref_token|>` are used by the paper's pairwise trajectory preference objective. `<|reward_token|>` and `<|sim_token|>` are preserved for checkpoint compatibility.
+
+The LeRobot config stores a serialized `vlm_config` with the post-resize vocabulary so the model can reload from `config.json` without downloading the base Qwen weights first. For `Qwen/Qwen3-VL-4B-Instruct`, the tokenizer length is `151669`, and the five ROBOMETER tokens produce the checkpoint vocabulary size `151674`.
+
+### Progress Prediction
+
+In the published checkpoint, progress is discrete. The progress head outputs logits over `progress_discrete_bins=10` uniformly spaced bin centers in `[0, 1]`. LeRobot converts these logits into a continuous value by applying a softmax and taking the expectation over bin centers, matching the upstream ROBOMETER implementation.
+
+### Success Prediction
+
+The success head outputs raw logits per frame. LeRobot converts them to probabilities with `sigmoid`. When `reward_output="success"`, `compute_reward()` thresholds the last-frame success probability using `success_threshold`.
+
+## Limitations
+
+- The current LeRobot integration is inference-only; it does not implement ROBOMETER training or preference-pair training.
+- `compute_reward()` returns a scalar per sample for the LeRobot reward-model API, even though ROBOMETER predicts per-frame progress and success internally.
+- ROBOMETER is video-language based; it does not use privileged robot state such as contact forces or object poses.
+
+## References
+
+- [ROBOMETER project](https://robometer.github.io/)
+- [ROBOMETER paper](https://arxiv.org/abs/2603.02115)
+- [Original ROBOMETER code](https://github.com/robometer/robometer)
+- [Published ROBOMETER-4B checkpoint](https://huggingface.co/lerobot/Robometer-4B)
+- [Qwen3-VL-4B-Instruct](https://huggingface.co/Qwen/Qwen3-VL-4B-Instruct)
+
+## Citation
+
+```bibtex
+@inproceedings{liang2026robometer,
+title = {Robometer: Scaling General-Purpose Robotic Reward Models via Trajectory Comparisons},
+author={Anthony Liang and Yigit Korkmaz and Jiahui Zhang and Minyoung Hwang and Abrar Anwar and Sidhant Kaushik and Aditya Shah and Alex S. Huang and Luke Zettlemoyer and Dieter Fox and Yu Xiang and Anqi Li and Andreea Bobu and Abhishek Gupta and Stephen Tu and Erdem Biyik and Jesse Zhang},
+year={2026},
+booktitle={Robotics: Science and Systems 2026},
+}
+```
+
+## License
+
+This LeRobot integration follows the **Apache 2.0 License** used by LeRobot. Check the upstream ROBOMETER code and model pages for the licenses of the original implementation and released checkpoints.

docs/source/sarm.mdx

@@ -46,7 +46,7 @@ This ensures identical task states map to consistent progress values, even acros
 
 ## Inputs and Targets (What the new code expects)
 
-SARM is trained through its processor (`src/lerobot/policies/sarm/processor_sarm.py`), which:
+SARM is trained through its processor (`src/lerobot/rewards/sarm/processor_sarm.py`), which:
 
 - **Encodes** images and task text with CLIP (ViT-B/32) into `video_features` and `text_features`
 - **Pads/truncates** robot state into `state_features` (up to `max_state_dim`)
@@ -347,7 +347,7 @@ Use `compute_rabc_weights.py` with `--visualize-only` to visualize model predict
 <hfoption id="single_stage">
 
 ```bash
-python src/lerobot/policies/sarm/compute_rabc_weights.py \
+python -m lerobot.rewards.sarm.compute_rabc_weights \
   --dataset-repo-id your-username/your-dataset \
   --reward-model-path your-username/sarm-model \
   --visualize-only \
@@ -360,7 +360,7 @@ python src/lerobot/policies/sarm/compute_rabc_weights.py \
 <hfoption id="dense_only">
 
 ```bash
-python src/lerobot/policies/sarm/compute_rabc_weights.py \
+python -m lerobot.rewards.sarm.compute_rabc_weights \
   --dataset-repo-id your-username/your-dataset \
   --reward-model-path your-username/sarm-model \
   --visualize-only \
@@ -373,7 +373,7 @@ python src/lerobot/policies/sarm/compute_rabc_weights.py \
 <hfoption id="dual">
 
 ```bash
-python src/lerobot/policies/sarm/compute_rabc_weights.py \
+python -m lerobot.rewards.sarm.compute_rabc_weights \
   --dataset-repo-id your-username/your-dataset \
   --reward-model-path your-username/sarm-model \
   --visualize-only \
@@ -429,7 +429,7 @@ The weighting follows **Equations 8-9** from the paper:
 First, run the SARM model on all frames in your dataset to compute progress values:
 
 ```bash
-python src/lerobot/policies/sarm/compute_rabc_weights.py \
+python -m lerobot.rewards.sarm.compute_rabc_weights \
   --dataset-repo-id your-username/your-dataset \
   --reward-model-path your-username/sarm-model \
   --head-mode sparse \
@@ -465,15 +465,15 @@ This script:
 
 ### Step 5b: Train Policy with RA-BC
 
-Once you have the progress file, train your policy with RA-BC weighting. The progress file is auto-detected from the dataset path (`sarm_progress.parquet`). Currently PI0, PI0.5 and SmolVLA are supported with RA-BC:
+Once you have the progress file, train your policy with RA-BC weighting. The progress file is auto-detected from the dataset path (`sarm_progress.parquet`) if not explicitly provided. Currently PI0, PI0.5 and SmolVLA are supported with RA-BC:
 
 ```bash
 lerobot-train \
   --dataset.repo_id=your-username/your-dataset \
   --policy.type=pi0 \
-  --use_rabc=true \
-  --rabc_head_mode=sparse \
-  --rabc_kappa=0.01 \
+  --sample_weighting.type=rabc \
+  --sample_weighting.head_mode=sparse \
+  --sample_weighting.kappa=0.01 \
   --output_dir=outputs/train/policy_rabc \
   --batch_size=32 \
   --steps=40000
@@ -488,12 +488,13 @@ The training script automatically:
 
 **RA-BC Arguments:**
 
-| Argument               | Description                                                | Default                            |
-| ---------------------- | ---------------------------------------------------------- | ---------------------------------- |
-| `--use_rabc`           | Enable RA-BC sample weighting                              | `false`                            |
-| `--rabc_progress_path` | Path to progress parquet file (auto-detected from dataset) | `sarm_progress.parquet` in dataset |
-| `--rabc_head_mode`     | Which SARM head's progress to use: `sparse` or `dense`     | `sparse`                           |
-| `--rabc_kappa`         | Threshold κ for high-quality samples                       | `0.01`                             |
+| Argument                           | Description                                            | Default                 |
+| ---------------------------------- | ------------------------------------------------------ | ----------------------- |
+| `--sample_weighting.type`          | Weighting strategy type (`rabc` or `uniform`)          | `rabc`                  |
+| `--sample_weighting.progress_path` | Path to progress parquet file                          | `sarm_progress.parquet` |
+| `--sample_weighting.head_mode`     | Which SARM head's progress to use: `sparse` or `dense` | `sparse`                |
+| `--sample_weighting.kappa`         | Threshold κ for high-quality samples                   | `0.01`                  |
+| `--sample_weighting.epsilon`       | Small constant for numerical stability                 | `1e-6`                  |
 
 ### Tuning RA-BC Kappa
 
@@ -511,30 +512,30 @@ The `kappa` parameter is the threshold that determines which samples get full we
 
 Monitor these WandB metrics during training:
 
-| Metric             | Healthy Range | Problem Indicator         |
-| ------------------ | ------------- | ------------------------- |
-| `rabc_mean_weight` | 0.3 - 0.8     | ≈ 1.0 means kappa too low |
-| `rabc_delta_mean`  | > 0           | Should be positive        |
-| `rabc_delta_std`   | > 0           | Variance in data quality  |
+| Metric                        | Healthy Range | Problem Indicator         |
+| ----------------------------- | ------------- | ------------------------- |
+| `sample_weight_mean_weight`   | 0.3 - 0.8     | ≈ 1.0 means kappa too low |
+| `sample_weighting/delta_mean` | > 0           | Should be positive        |
+| `sample_weighting/delta_std`  | > 0           | Variance in data quality  |
 
-**If `rabc_mean_weight ≈ 1.0`:** Your kappa is too low. Most samples have `delta > kappa` and bypass the soft-weighting entirely. RA-BC becomes equivalent to vanilla BC.
+**If `sample_weight_mean_weight ≈ 1.0`:** Your kappa is too low. Most samples have `delta > kappa` and bypass the soft-weighting entirely. RA-BC becomes equivalent to vanilla BC.
 
 **Setting kappa based on your data:**
 
-The default `kappa=0.01` was tuned for the paper's T-shirt folding task (~90s episodes at 30fps). For your dataset, check the logged `rabc_delta_mean` and `rabc_delta_std`:
+The default `kappa=0.01` was tuned for the paper's T-shirt folding task (~90s episodes at 30fps). For your dataset, check the logged `sample_weighting/delta_mean` and `sample_weighting/delta_std`:
 
 ```
 # If delta_mean ≈ 0.03 and delta_std ≈ 0.02:
 # Most deltas fall in range [0.01, 0.05]
 
 # Option 1: Set kappa = delta_mean (medium selectivity)
---rabc_kappa=0.03
+--sample_weighting.kappa=0.03
 
 # Option 2: Set kappa = delta_mean + delta_std (high selectivity)
---rabc_kappa=0.05
+--sample_weighting.kappa=0.05
 
 # Option 3: Set kappa = delta_mean + 2*delta_std (very selective)
---rabc_kappa=0.07
+--sample_weighting.kappa=0.07
 ```
 
 **When RA-BC may not help:**
@@ -550,8 +551,8 @@ accelerate launch \
   src/lerobot/scripts/lerobot_train.py \
   --dataset.repo_id=your-username/your-dataset \
   --policy.type=pi0 \
-  --use_rabc=true \
-  --rabc_kappa=0.01 \
+  --sample_weighting.type=rabc \
+  --sample_weighting.kappa=0.01 \
   --output_dir=outputs/train/policy_rabc \
   --batch_size=32 \
   --steps=40000
@@ -576,7 +577,7 @@ accelerate launch \
 ### RA-BC
 
 1. **Train SARM first**: RA-BC quality depends entirely on SARM quality
-2. **Monitor `rabc_mean_weight`**: If it's ≈ 1.0, increase kappa (see [Tuning RA-BC Kappa](#tuning-ra-bc-kappa))
+2. **Monitor `sample_weight_mean_weight`**: If it's ≈ 1.0, increase kappa (see [Tuning RA-BC Kappa](#tuning-ra-bc-kappa))
 
 ---
 

docs/source/smolvla.mdx

@@ -97,22 +97,22 @@ Similarly for when recording an episode, it is recommended that you are logged i
 Once you are logged in, you can run inference in your setup by doing:
 
 ```bash
-lerobot-record \
+lerobot-rollout \
+  --strategy.type=base \
   --robot.type=so101_follower \
   --robot.port=/dev/ttyACM0 \ # <- Use your port
   --robot.id=my_blue_follower_arm \ # <- Use your robot id
   --robot.cameras="{ front: {type: opencv, index_or_path: 8, width: 640, height: 480, fps: 30}}" \ # <- Use your cameras
-  --dataset.single_task="Grasp a lego block and put it in the bin." \ # <- Use the same task description you used in your dataset recording
-  --dataset.repo_id=${HF_USER}/eval_DATASET_NAME_test \  # <- This will be the dataset name on HF Hub
-  --dataset.episode_time_s=50 \
-  --dataset.num_episodes=10 \
-  --dataset.streaming_encoding=true \
-  --dataset.encoder_threads=2 \
-  # --dataset.vcodec=auto \
+  --task="Grasp a lego block and put it in the bin." \ # <- Use the same task description you used in your dataset recording
+  # <- RTC optional, use when running on low power hardware \
+  # --inference.type=rtc \
+  # --inference.rtc.execution_horizon=10 \
+  # --inference.rtc.max_guidance_weight=10.0 \
   # <- Teleop optional if you want to teleoperate in between episodes \
   # --teleop.type=so100_leader \
   # --teleop.port=/dev/ttyACM0 \
   # --teleop.id=my_red_leader_arm \
+  # --display_data=true #optional use if you want to see the camera stream \
   --policy.path=HF_USER/FINETUNE_MODEL_NAME # <- Use your fine-tuned model
 ```
 

docs/source/streaming_video_encoding.mdx

@@ -17,9 +17,9 @@ This makes `save_episode()` near-instant (the video is already encoded by the ti
 | Parameter               | CLI Flag                          | Type          | Default       | Description                                                       |
 | ----------------------- | --------------------------------- | ------------- | ------------- | ----------------------------------------------------------------- |
 | `streaming_encoding`    | `--dataset.streaming_encoding`    | `bool`        | `True`        | Enable real-time encoding during capture                          |
-| `vcodec`                | `--dataset.vcodec`                | `str`         | `"libsvtav1"` | Video codec. `"auto"` detects best HW encoder                     |
+| `vcodec`                | `--dataset.camera_encoder.vcodec` | `str`         | `"libsvtav1"` | Video codec. `"auto"` detects best HW encoder                     |
 | `encoder_threads`       | `--dataset.encoder_threads`       | `int \| None` | `None` (auto) | Threads per encoder instance. `None` will leave the vcoded decide |
-| `encoder_queue_maxsize` | `--dataset.encoder_queue_maxsize` | `int`         | `60`          | Max buffered frames per camera (~2s at 30fps). Consumes RAM       |
+| `encoder_queue_maxsize` | `--dataset.encoder_queue_maxsize` | `int`         | `30`          | Max buffered frames per camera (~1s at 30fps). Consumes RAM       |
 
 ## 3. Performance Considerations
 
@@ -48,7 +48,7 @@ This parameter controls how many threads each encoder instance uses internally:
 
 ### Backpressure and Frame Dropping
 
-Each camera has a bounded queue (`encoder_queue_maxsize`, default 60 frames). When the encoder can't keep up:
+Each camera has a bounded queue (`encoder_queue_maxsize`, default 30 frames). When the encoder can't keep up:
 
 1. The queue fills up (consuming RAM)
 2. New frames are **dropped** (not blocked) — the capture loop continues uninterrupted
@@ -82,15 +82,15 @@ Use HW encoding when:
 
 ### Available HW Encoders
 
-| Encoder             | Platform      | Hardware                                                                                         | CLI Value                            |
-| ------------------- | ------------- | ------------------------------------------------------------------------------------------------ | ------------------------------------ |
-| `h264_videotoolbox` | macOS         | Apple Silicon / Intel                                                                            | `--dataset.vcodec=h264_videotoolbox` |
-| `hevc_videotoolbox` | macOS         | Apple Silicon / Intel                                                                            | `--dataset.vcodec=hevc_videotoolbox` |
-| `h264_nvenc`        | Linux/Windows | NVIDIA GPU                                                                                       | `--dataset.vcodec=h264_nvenc`        |
-| `hevc_nvenc`        | Linux/Windows | NVIDIA GPU                                                                                       | `--dataset.vcodec=hevc_nvenc`        |
-| `h264_vaapi`        | Linux         | Intel/AMD GPU                                                                                    | `--dataset.vcodec=h264_vaapi`        |
-| `h264_qsv`          | Linux/Windows | Intel Quick Sync                                                                                 | `--dataset.vcodec=h264_qsv`          |
-| `auto`              | Any           | Probes the system for available HW encoders. Falls back to `libsvtav1` if no HW encoder is found | `--dataset.vcodec=auto`              |
+| Encoder             | Platform      | Hardware                                                                                         | CLI Value                                           |
+| ------------------- | ------------- | ------------------------------------------------------------------------------------------------ | --------------------------------------------------- |
+| `h264_videotoolbox` | macOS         | Apple Silicon / Intel                                                                            | `--dataset.camera_encoder.vcodec=h264_videotoolbox` |
+| `hevc_videotoolbox` | macOS         | Apple Silicon / Intel                                                                            | `--dataset.camera_encoder.vcodec=hevc_videotoolbox` |
+| `h264_nvenc`        | Linux/Windows | NVIDIA GPU                                                                                       | `--dataset.camera_encoder.vcodec=h264_nvenc`        |
+| `hevc_nvenc`        | Linux/Windows | NVIDIA GPU                                                                                       | `--dataset.camera_encoder.vcodec=hevc_nvenc`        |
+| `h264_vaapi`        | Linux         | Intel/AMD GPU                                                                                    | `--dataset.camera_encoder.vcodec=h264_vaapi`        |
+| `h264_qsv`          | Linux/Windows | Intel Quick Sync                                                                                 | `--dataset.camera_encoder.vcodec=h264_qsv`          |
+| `auto`              | Any           | Probes the system for available HW encoders. Falls back to `libsvtav1` if no HW encoder is found | `--dataset.camera_encoder.vcodec=auto`              |
 
 > [!NOTE]
 > In order to use the HW accelerated encoders you might need to upgrade your GPU drivers.
@@ -100,15 +100,15 @@ Use HW encoding when:
 
 ## 5. Troubleshooting
 
-| Symptom                                                            | Likely Cause                                 | Fix                                                                                                                                                                                                                                                                                  |
-| ------------------------------------------------------------------ | -------------------------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ |
-| System freezes or choppy robot movement or Rerun visualization lag | CPU starved (100% load usage)                | Close other apps, reduce encoding throughput, lower `encoder_threads`, use `h264`, use `display_data=False`. If the CPU continues to be at 100% then it might be insufficient for your setup, consider `--dataset.streaming_encoding=false` or HW encoding (`--dataset.vcodec=auto`) |
-| "Encoder queue full" warnings or dropped frames in dataset         | Encoder can't keep up (Queue overflow)       | If CPU is not at 100%: Increase `encoder_threads`, increase `encoder_queue_maxsize` or use HW encoding (`--dataset.vcodec=auto`).                                                                                                                                                    |
-| High RAM usage                                                     | Queue filling faster than encoding           | `encoder_threads` too low or CPU insufficient. Reduce `encoder_queue_maxsize` or use HW encoding                                                                                                                                                                                     |
-| Large video files                                                  | Using HW encoder or H.264                    | Expected trade-off. Switch to `libsvtav1` if CPU allows                                                                                                                                                                                                                              |
-| `save_episode()` still slow                                        | `streaming_encoding` is `False`              | Set `--dataset.streaming_encoding=true`                                                                                                                                                                                                                                              |
-| Encoder thread crash                                               | Codec not available or invalid settings      | Check `vcodec` is installed, try `--dataset.vcodec=auto`                                                                                                                                                                                                                             |
-| Recorded dataset is missing frames                                 | CPU/GPU starvation or occasional load spikes | If ~5% of frames are missing, your system is likely overloaded — follow the recommendations above. If fewer frames are missing (~2%), they are probably due to occasional transient load spikes (often at startup) and can be considered expected.                                   |
+| Symptom                                                            | Likely Cause                                 | Fix                                                                                                                                                                                                                                                                                                 |
+| ------------------------------------------------------------------ | -------------------------------------------- | --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
+| System freezes or choppy robot movement or Rerun visualization lag | CPU starved (100% load usage)                | Close other apps, reduce encoding throughput, lower `encoder_threads`, use `h264`, use `display_data=False`. If the CPU continues to be at 100% then it might be insufficient for your setup, consider `--dataset.streaming_encoding=false` or HW encoding (`--dataset.camera_encoder.vcodec=auto`) |
+| "Encoder queue full" warnings or dropped frames in dataset         | Encoder can't keep up (Queue overflow)       | If CPU is not at 100%: Increase `encoder_threads`, increase `encoder_queue_maxsize` or use HW encoding (`--dataset.camera_encoder.vcodec=auto`).                                                                                                                                                    |
+| High RAM usage                                                     | Queue filling faster than encoding           | `encoder_threads` too low or CPU insufficient. Reduce `encoder_queue_maxsize` or use HW encoding                                                                                                                                                                                                    |
+| Large video files                                                  | Using HW encoder or H.264                    | Expected trade-off. Switch to `libsvtav1` if CPU allows                                                                                                                                                                                                                                             |
+| `save_episode()` still slow                                        | `streaming_encoding` is `False`              | Set `--dataset.streaming_encoding=true`                                                                                                                                                                                                                                                             |
+| Encoder thread crash                                               | Codec not available or invalid settings      | Check `vcodec` is installed, try `--dataset.camera_encoder.vcodec=auto`                                                                                                                                                                                                                             |
+| Recorded dataset is missing frames                                 | CPU/GPU starvation or occasional load spikes | If ~5% of frames are missing, your system is likely overloaded — follow the recommendations above. If fewer frames are missing (~2%), they are probably due to occasional transient load spikes (often at startup) and can be considered expected.                                                  |
 
 ## 6. Recommended Configurations
 
@@ -146,7 +146,7 @@ On very constrained systems, streaming encoding may compete too heavily with the
 # 2camsx 640x480x3 @30fps: Requires some tuning.
 
 # Use H.264, disable streaming, consider batching encoding
-lerobot-record --dataset.vcodec=h264 --dataset.streaming_encoding=false ...
+lerobot-record --dataset.camera_encoder.vcodec=h264 --dataset.streaming_encoding=false ...
 ```
 
 ## 7. Closing note

docs/source/tools.mdx

@@ -0,0 +1,210 @@
+# Tools
+
+LeRobot v3.1 supports **tool calls** in policies — assistant messages can
+emit structured invocations like `say(text="OK, starting now")` that the
+runtime dispatches to a real implementation (TTS, controller, logger, …).
+
+This page covers:
+
+1. Where the tool catalog lives.
+2. How the annotation pipeline produces tool-call atoms.
+3. How to add your own tool.
+
+## Where tools are declared
+
+Two layers.
+
+**The catalog** — a list of OpenAI-style function schemas — lives at
+`meta/info.json["tools"]` on each dataset. Example:
+
+```json
+{
+  "features": { "...": "..." },
+  "tools": [
+    {
+      "type": "function",
+      "function": {
+        "name": "say",
+        "description": "Speak a short utterance to the user via the TTS executor.",
+        "parameters": {
+          "type": "object",
+          "properties": {
+            "text": {
+              "type": "string",
+              "description": "The verbatim text to speak."
+            }
+          },
+          "required": ["text"]
+        }
+      }
+    }
+  ]
+}
+```
+
+Read it via the dataset metadata accessor:
+
+```python
+from lerobot.datasets.dataset_metadata import LeRobotDatasetMetadata
+
+meta = LeRobotDatasetMetadata(repo_id="pepijn/super_poulain_final_annotations")
+tools = meta.tools     # list[dict] — OpenAI tool schemas
+```
+
+If the dataset's `info.json` doesn't declare any tools, `meta.tools`
+returns `DEFAULT_TOOLS` from `lerobot.datasets.language` — currently a
+single-entry list with the canonical `say` schema. So unannotated
+datasets and chat-template consumers keep working without any
+configuration:
+
+```python
+prompt_str = tokenizer.apply_chat_template(
+    sample["messages"],
+    tools=meta.tools,                 # works either way
+    add_generation_prompt=False,
+    tokenize=False,
+)
+```
+
+**The implementations** — runnable Python — will live under
+`src/lerobot/tools/`, one file per tool. The runtime dispatcher and
+the canonical `say` implementation (wrapping Kyutai's pocket-tts) are
+not part of the catalog layer described here; today this layer ships
+only the schema storage and the `DEFAULT_TOOLS` fallback constant.
+
+## Per-row tool _invocations_
+
+The catalog above describes _what can be called_. The actual _call_ — the
+function name plus the argument values — is stored per-row, on the
+assistant atoms in `language_events`:
+
+```python
+{
+  "role": "assistant",
+  "content": null,
+  "style": null,
+  "timestamp": 12.4,
+  "camera": null,
+  "tool_calls": [
+    { "type": "function",
+      "function": { "name": "say", "arguments": { "text": "On it." } } }
+  ]
+}
+```
+
+Recipes splice these into rendered messages via `tool_calls_from`:
+
+```yaml
+user_interjection_response:
+  bindings:
+    speech: "emitted_at(t, role=assistant, tool_name=say)"
+  messages:
+    - { role: user, content: "${task}", stream: high_level }
+    - {
+        role: assistant,
+        content: "${current_plan}",
+        stream: high_level,
+        target: true,
+        tool_calls_from: speech,
+      }
+```
+
+The model's training target is one assistant turn that carries both the
+plan text _and_ the `say` tool call. At inference, the runtime parses
+the generated text back into structured `tool_calls` and dispatches to
+the matching implementation.
+
+## How to add your own tool
+
+> **Note:** Steps 2 and 3 below describe the runtime layer
+> (`src/lerobot/tools/`, the `Tool` protocol, `TOOL_REGISTRY`,
+> `get_tools(meta)`) which is not part of the catalog layer shipped
+> today — those modules don't yet exist in the tree. Step 1 alone is
+> enough to make the tool visible to the chat template via
+> `meta.tools` so the model can learn to _generate_ the call;
+> executing the call at inference requires the runtime layer.
+
+Three steps. Concrete example: a `record_observation` tool the policy
+can call to capture an extra observation outside the regular control
+loop.
+
+### Step 1 — declare the schema
+
+Add an entry under `meta/info.json["tools"]`. Either edit the file
+directly on disk _before_ running the annotation pipeline (it'll be
+preserved) or hand it to `lerobot-annotate` via a config flag.
+
+```json
+{
+  "tools": [
+    { "type": "function", "function": { "name": "say", "...": "..." } },
+    {
+      "type": "function",
+      "function": {
+        "name": "record_observation",
+        "description": "Capture a high-resolution still image for the user.",
+        "parameters": {
+          "type": "object",
+          "properties": {
+            "label": {
+              "type": "string",
+              "description": "Short label for the saved image."
+            }
+          },
+          "required": ["label"]
+        }
+      }
+    }
+  ]
+}
+```
+
+The schema follows OpenAI's function-calling convention exactly, so the
+chat template can render it natively.
+
+### Step 2 — implement the call
+
+Create `src/lerobot/tools/record_observation.py`:
+
+```python
+from .base import Tool
+from typing import Any
+
+RECORD_OBSERVATION_SCHEMA: dict[str, Any] = { "...": "..." }   # mirrors the JSON above
+
+
+class RecordObservationTool:
+    name = "record_observation"
+    schema = RECORD_OBSERVATION_SCHEMA
+
+    def __init__(self, schema: dict | None = None, output_dir: str = "."):
+        self.output_dir = output_dir
+
+    def call(self, arguments: dict) -> str:
+        label = arguments["label"]
+        # ... save the latest camera frame to <output_dir>/<label>.png ...
+        return f"saved {label}.png"
+```
+
+One file per tool keeps dependencies isolated — `record_observation`
+might pull `pillow`, while `say` pulls `pocket-tts`. Users installing
+only the tools they need avoid heavy transitive deps.
+
+### Step 3 — register it
+
+Add to `src/lerobot/tools/registry.py`:
+
+```python
+from .record_observation import RecordObservationTool
+
+TOOL_REGISTRY["record_observation"] = RecordObservationTool
+```
+
+That's it. At runtime `get_tools(meta)` looks up each schema in
+`meta.tools`, instantiates the matching registered class, and returns
+a name → instance dict the dispatcher can route into.
+
+If you want to use a tool _without_ writing an implementation (e.g. for
+training-time chat-template formatting only), step 1 alone is enough —
+the model still learns to _generate_ the call. Steps 2 and 3 are only
+needed to actually _execute_ it at inference.

docs/source/topreward.mdx

@@ -0,0 +1,177 @@
+# TOPReward
+
+TOPReward is a **zero-shot reward model** that extracts token log-probabilities from an off-the-shelf vision-language model (VLM) as a robotic reward signal. Given a video trajectory and a task instruction, it returns the VLM's log-likelihood that the instruction is true — no fine-tuning required.
+
+**Paper**: [TOPReward: Token Probabilities as Hidden Zero-Shot Rewards for Robotics](https://arxiv.org/abs/2602.19313)
+**Project**: [topreward.github.io](https://topreward.github.io/webpage/)
+**Original code**: [github.com/TOPReward/TOPReward](https://github.com/TOPReward/TOPReward)
+**Default backbone**: [Qwen/Qwen3-VL-8B-Instruct](https://huggingface.co/Qwen/Qwen3-VL-8B-Instruct)
+
+## Overview
+
+TOPReward asks a generic VLM how likely a task instruction is, **conditioned on the video** of a robot trying to complete that task. Concretely, given:
+
+- A trajectory video (a sequence of frames).
+- A task instruction (e.g. _"open the drawer"_).
+
+it builds a chat prompt of the form
+
+```text
+<video>
+"The above video shows a robot manipulation trajectory that completes the
+ following task: <instruction> Decide whether the above statement is True
+ or not. The answer is: True"
+```
+
+forwards it through the VLM, label-masks everything except the very last token, and reads back the log-probability of that token — by default the literal `"True"` that closes the suffix template. The resulting `log P("True" | video + prompt + instruction)` is the reward.
+
+Because the method only depends on a frozen VLM, TOPReward is **zero-shot**: there are no fine-tuned weights to host. The "model" in LeRobot is a small wrapper around `transformers`' `Qwen3VLForConditionalGeneration` plus the label-masking logic. The processor owns the tokeniser and builds the full chat prompt (EO-1/Robometer pattern).
+
+## What the LeRobot integration covers
+
+- Standard `reward_model.type=topreward` configuration through LeRobot.
+- VLM loading via the `transformers` `Qwen3VLForConditionalGeneration` API.
+- Prompt assembly + tokenisation in the processor (matching upstream `QwenClient.compute_instruction_reward`).
+- `compute_reward()` returns one scalar log-prob per sample.
+- LeRobot reward-model save/load — `save_pretrained` writes only `config.json` (the VLM is identified by `vlm_name`).
+- An offline labeling script that writes a `topreward_progress.parquet` (SARM-compatible schema) for RA-BC and overlay.
+
+The current LeRobot port supports the **Qwen3-VL client only**. Other upstream clients (Gemini, OpenAI, Gemma, Molmo) can be added as follow-up extras.
+
+## Installation Requirements
+
+1. Install LeRobot following the [Installation Guide](./installation).
+2. Install the TOPReward optional extra:
+
+```bash
+pip install -e ".[topreward]"
+```
+
+or, with `uv` from a source checkout:
+
+```bash
+uv sync --extra topreward
+```
+
+This pulls in `transformers`. The first time you run TOPReward, Hugging Face will also download the VLM weights from the Hub (~16 GB for Qwen3-VL-8B-Instruct). A GPU is strongly recommended.
+
+## Model Inputs and Outputs
+
+TOPReward expects:
+
+- A trajectory video or sequence of frames.
+- A natural-language task description.
+
+In LeRobot datasets the preprocessor reads:
+
+| Config field              | Default                     | Meaning                                       |
+| ------------------------- | --------------------------- | --------------------------------------------- |
+| `reward_model.image_key`  | `observation.images.top`    | Camera observation used by TOPReward          |
+| `reward_model.task_key`   | `task`                      | Key in complementary data for the task string |
+| `reward_model.max_frames` | `16`                        | Cap on frames per sample                      |
+| `reward_model.fps`        | `2.0`                       | Metadata passed to the Qwen video processor   |
+| `reward_model.vlm_name`   | `Qwen/Qwen3-VL-8B-Instruct` | Hugging Face Hub id of the underlying VLM     |
+
+The model returns:
+
+- `compute_reward(batch)`: one log-probability per sample. Higher = better task-video alignment. When `success_threshold` is finite, returns the binary thresholded value instead.
+
+## Usage
+
+### Load the reward model directly
+
+```python
+from lerobot.rewards.topreward import TOPRewardConfig, TOPRewardModel
+
+cfg = TOPRewardConfig(
+    vlm_name="Qwen/Qwen3-VL-8B-Instruct",
+    device="cuda",
+)
+reward_model = TOPRewardModel(cfg)
+```
+
+### Use the reward factory
+
+```python
+from lerobot.rewards import make_reward_model, make_reward_model_config, make_reward_pre_post_processors
+
+cfg = make_reward_model_config(
+    "topreward",
+    vlm_name="Qwen/Qwen3-VL-8B-Instruct",
+    device="cuda",
+    image_key="observation.images.top",
+)
+reward_model = make_reward_model(cfg)
+preprocessor, postprocessor = make_reward_pre_post_processors(cfg)
+```
+
+The preprocessor tokenises the full prompt (video + prefix + instruction suffix), writes Qwen-VL tensors + `prompt_length` under `observation.topreward.*`. The model reads those tensors, label-masks based on `prompt_length`, and extracts the log-prob reward.
+
+### Offline dataset labeling
+
+Write a `topreward_progress.parquet` for RA-BC training and overlay videos:
+
+```bash
+# Sparse-dense (15 anchors per episode, matches upstream)
+uv run python -m lerobot.rewards.topreward.compute_rabc_weights \
+    --dataset-repo-id lerobot/libero_10_image \
+    --num-samples 15 \
+    --device cuda
+```
+
+Then render the progress overlay for any episode:
+
+```bash
+uv run examples/dataset/create_progress_videos.py \
+    --repo-id lerobot/libero_10_image \
+    --episode 0 \
+    --progress-file topreward_progress.parquet \
+    --gif
+```
+
+## Configuration Notes
+
+### Prompt knobs
+
+The default prompt mirrors the upstream paper:
+
+```text
+prompt_prefix = "The above video shows a robot manipulation trajectory that completes the following task: "
+prompt_suffix_template = "{instruction} Decide whether the above statement is True or not. The answer is: True"
+```
+
+Both are exposed on `TOPRewardConfig` for ablation. The suffix template **must** contain `{instruction}`.
+
+### Chat template
+
+`add_chat_template=True` wraps the full prompt (including instruction) with the tokenizer's chat template before tokenisation. Default is `False`, matching the upstream paper's main experiments.
+
+## Limitations
+
+- The current LeRobot port is **inference-only and zero-shot**; `forward()` is not overridden and `is_trainable` returns `False`.
+- Only the **Qwen3-VL family** is supported; other upstream clients are out of scope.
+- TOPReward inherits the underlying VLM's biases.
+
+## References
+
+- [TOPReward project page](https://topreward.github.io/webpage/)
+- [TOPReward paper](https://arxiv.org/abs/2602.19313)
+- [Original TOPReward code](https://github.com/TOPReward/TOPReward)
+- [Qwen3-VL-8B-Instruct](https://huggingface.co/Qwen/Qwen3-VL-8B-Instruct)
+
+## Citation
+
+```bibtex
+@article{chen2026topreward,
+  title={TOPReward: Token Probabilities as Hidden Zero-Shot Rewards for Robotics},
+  author={Chen, Shirui and Harrison, Cole and Lee, Ying-Chun and Yang, Angela Jin and
+          Ren, Zhongzheng and Ratliff, Lillian J and Duan, Jiafei and Fox, Dieter and
+          Krishna, Ranjay},
+  journal={arXiv preprint arXiv:2602.19313},
+  year={2026}
+}
+```
+
+## License
+
+The original TOPReward codebase is MIT-licensed. The LeRobot port follows the LeRobot Apache 2.0 license; the wrapped Qwen3-VL weights are subject to the original Qwen license.

docs/source/using_dataset_tools.mdx

@@ -117,10 +117,10 @@ lerobot-edit-dataset \
     --repo_id lerobot/pusht_image \
     --operation.type convert_image_to_video \
     --operation.output_dir outputs/pusht_video \
-    --operation.vcodec libsvtav1 \
-    --operation.pix_fmt yuv420p \
-    --operation.g 2 \
-    --operation.crf 30
+    --operation.camera_encoder.vcodec libsvtav1 \
+    --operation.camera_encoder.pix_fmt yuv420p \
+    --operation.camera_encoder.g 2 \
+    --operation.camera_encoder.crf 30
 
 # Convert only specific episodes
 lerobot-edit-dataset \
@@ -147,11 +147,7 @@ lerobot-edit-dataset \
 **Parameters:**
 
 - `output_dir`: Custom output directory (optional - by default uses `new_repo_id` or `{repo_id}_video`)
-- `vcodec`: Video codec to use - options: `h264`, `hevc`, `libsvtav1` (default: `libsvtav1`)
-- `pix_fmt`: Pixel format - options: `yuv420p`, `yuv444p` (default: `yuv420p`)
-- `g`: Group of pictures (GOP) size - lower values give better quality but larger files (default: 2)
-- `crf`: Constant rate factor - lower values give better quality but larger files, 0 is lossless (default: 30)
-- `fast_decode`: Fast decode tuning option (default: 0)
+- `camera_encoder`: Video encoder settings — all sub-fields accessible via `--operation.camera_encoder.<field>. See [Video Encoding Parameters](./video_encoding_parameters) for more details.
 - `episode_indices`: List of specific episodes to convert (default: all episodes)
 - `num_workers`: Number of parallel workers for processing (default: 4)
 

docs/source/video_encoding_parameters.mdx

@@ -0,0 +1,117 @@
+# Video encoding parameters
+
+When video storage is enabled, LeRobot stores each camera stream as an **MP4** file instead of saving one image file per timestep. Video encoding compresses across time, which usually cuts dataset size and I/O compared to a pile of PNG, while keeping MP4 — a format every player and loader understands.
+
+Encoding frames into an MP4 is a full FFmpeg pipeline: choice of encoder, pixel format, GOP/keyframes, quality vs. speed, and optional extra encoder flags. Most of these knobs are user-tunable through `camera_encoder`, a nested `VideoEncoderConfig` (`lerobot.configs.video.VideoEncoderConfig`) passed through PyAV.
+
+You can set these parameters from the CLI with `--dataset.camera_encoder.<field>` (e.g. with `lerobot-record` or `lerobot-rollout`). The same block applies to every camera video stream in that run.
+
+<Tip>
+  Video storage must be on for `camera_encoder` to have any effect —
+  `use_videos=True` in Python APIs, or `--dataset.video=true` on the CLI (the
+  recording default). With video off, inputs stay as images and `camera_encoder`
+  is ignored.
+</Tip>
+
+For details on **when** frames are written vs. encoded (streaming vs. post-episode), queues, and other top-level `--dataset.*` switches, see [Streaming Video Encoding](./streaming_video_encoding). For an encoding-parameter comparison and experiments, see the [video-benchmark Space](https://huggingface.co/spaces/lerobot/video-benchmark).
+
+---
+
+## Example
+
+```bash
+lerobot-record \
+    --robot.type=so100_follower \
+    --robot.port=/dev/tty.usbmodem58760431541 \
+    --robot.cameras="{laptop: {type: opencv, index_or_path: 0, width: 640, height: 480, fps: 30}}" \
+    --robot.id=black \
+    --teleop.type=so100_leader \
+    --teleop.port=/dev/tty.usbmodem58760431551 \
+    --teleop.id=blue \
+    --dataset.repo_id=<my_username>/<my_dataset_name> \
+    --dataset.num_episodes=2 \
+    --dataset.single_task="Grab the cube" \
+    --dataset.streaming_encoding=true \
+    --dataset.encoder_threads=2 \
+    --dataset.camera_encoder.vcodec=h264 \
+    --dataset.camera_encoder.preset=fast \
+    --dataset.camera_encoder.extra_options={"tune": "film", "profile:v": "high", "bf": 2} \
+    --display_data=true
+```
+
+---
+
+## Tuning parameters
+
+<Tip warning={true}>
+The defaults are tuned to balance **compression ratio**, **visual quality**, and **decoding/seek speed** for typical robotics datasets. Changing them can affect both recording (CPU load, frame drops) and training (decoding throughput, image quality).
+
+Only override these parameters if you have a specific reason to, and measure the impact on your pipeline before relying on the new settings.
+
+</Tip>
+
+All flags below are prefixed with `--dataset.camera_encoder.` on the CLI.
+
+| Parameter       | Type             | Default       | Description                                                                                                                                                                            |
+| --------------- | ---------------- | ------------- | -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
+| `vcodec`        | `str`            | `"libsvtav1"` | Video codec name. `"auto"` picks the first available hardware encoder from a fixed preference list, falling back to `libsvtav1`.                                                       |
+| `pix_fmt`       | `str`            | `"yuv420p"`   | Output pixel format. Must be supported by the chosen codec in your FFmpeg build.                                                                                                       |
+| `g`             | `int`            | `2`           | GOP size — a keyframe every `g` frames. Emitted as FFmpeg option `g`.                                                                                                                  |
+| `crf`           | `int` or `float` | `30`          | Abstract quality value, mapped per codec (see the [mapping](#mapping-videoencoderconfig--ffmpeg-options) below). Lower → higher quality / larger output where the mapping is monotone. |
+| `preset`        | `int` or `str`   | `12` \*       | Encoder speed preset; meaning depends on the codec. <br/>\* When unset and `vcodec=libsvtav1`, LeRobot defaults to `12`.                                                               |
+| `fast_decode`   | `int`            | `0`           | `libsvtav1`: `0–2`, passed via `svtav1-params`. <br/>`h264` / `hevc` (software): if `>0`, sets `tune=fastdecode`. <br/>Other codecs: usually unused.                                   |
+| `video_backend` | `str`            | `"pyav"`      | Only `"pyav"` is currently implemented for video encoding.                                                                                                                             |
+| `extra_options` | `dict`           | `{}`          | Extra FFmpeg or codec specific options merged after the structured fields above. Cannot override keys already set by those fields.                                                     |
+
+---
+
+## Persistence in dataset metadata
+
+After the first episode of a video stream is encoded, the encoder configuration is **persisted into the dataset metadata** (`meta/info.json`) under each video feature, alongside the values probed from the file itself. For a video feature `observation.images.<camera>`, the layout in `info.json` is:
+
+```json
+{
+  "features": {
+    "observation.images.laptop": {
+      "dtype": "video",
+      "shape": [480, 640, 3],
+      "info": {
+        "video.height": 480,
+        "video.width": 640,
+        "video.codec": "h264",
+        "video.pix_fmt": "yuv420p",
+        "video.fps": 30,
+        "video.channels": 3,
+        "video.is_depth_map": false,
+        "video.g": 2,
+        "video.crf": 30,
+        "video.preset": "fast",
+        "video.fast_decode": 0,
+        "video.video_backend": "pyav",
+        "video.extra_options": { "tune": "film", "profile:v": "high", "bf": 2 }
+      }
+    }
+  }
+}
+```
+
+Two sources contribute to the `info` block:
+
+- **Stream-derived** (read back from the encoded MP4 with PyAV): `video.height`, `video.width`, `video.codec`, `video.pix_fmt`, `video.fps`, `video.channels`, `video.is_depth_map`, plus `audio.*` if an audio stream is present.
+- **Encoder-derived** (taken from `VideoEncoderConfig`): `video.g`, `video.crf`, `video.preset`, `video.fast_decode`, `video.video_backend`, `video.extra_options`.
+
+<Tip>
+  This block is populated **once**, from the **first** episode. It assumes every
+  episode in the dataset was encoded with the same `camera_encoder`. Changing
+  encoder settings partway through a recording is not supported — the
+  `info.json` will only reflect the parameters used for the first episode.
+</Tip>
+
+---
+
+## Merging datasets
+
+When aggregating datasets with `merge_datasets`, video files are concatenated as-is (no re-encoding), and encoder fields in `info.json` are merged per-key:
+
+- **Stream-derived fields must match** across sources: `video.codec`, `video.pix_fmt`, `video.height`, `video.width`, `video.fps`. Otherwise FFmpeg's concat demuxer fails.
+- **Encoder-tuning fields are merged loosely**: `video.g`, `video.crf`, `video.preset`, `video.fast_decode`, `video.extra_options`. If every source agrees, the value is kept; if not, it's set to `null` (or `{}` for `video.extra_options`) and a warning is logged.

docs/source/xvla.mdx

@@ -220,7 +220,7 @@ REAL_DIM = 12
 # Postprocessing: Trim 20D predictions to 12D for deployment
 ```
 
-See the [action_hub.py](/home/jade_choghari/robot/lerobot/src/lerobot/policies/xvla/action_hub.py) implementation for details.
+See the [action_hub.py](https://github.com/huggingface/lerobot/blob/main/src/lerobot/policies/xvla/action_hub.py) implementation for details.
 
 #### Auto Action Mode (Recommended)
 
@@ -519,9 +519,9 @@ If you use X-VLA in your research, please cite:
 
 - [X-VLA Paper](https://arxiv.org/pdf/2510.10274)
 - [LeRobot Documentation](https://github.com/huggingface/lerobot)
-- [Action Registry Implementation](https://github.com/huggingface/lerobot/src/lerobot/policies/xvla/action_hub.py)
-- [Processor Implementation](https://github.com/huggingface/lerobot/src/lerobot/policies/xvla/processor_xvla.py)
-- [Model Configuration](https://github.com/huggingface/lerobot/src/lerobot/policies/xvla/configuration_xvla.py)
+- [Action Registry Implementation](https://github.com/huggingface/lerobot/blob/main/src/lerobot/policies/xvla/action_hub.py)
+- [Processor Implementation](https://github.com/huggingface/lerobot/blob/main/src/lerobot/policies/xvla/processor_xvla.py)
+- [Model Configuration](https://github.com/huggingface/lerobot/blob/main/src/lerobot/policies/xvla/configuration_xvla.py)
 
 ## Contributing
 

examples/annotations/run_hf_job.py

@@ -0,0 +1,119 @@
+#!/usr/bin/env python
+"""Launch ``lerobot-annotate`` on a Hugging Face job (vllm + Qwen3.6-27B VLM).
+
+Spawns one ``h200x4`` job that:
+
+  1. installs this branch of ``lerobot`` plus the annotation extras,
+  2. boots four vllm servers (one per GPU) with Qwen3.6-27B (dense VLM),
+  3. runs the plan / interjections / vqa modules across the dataset
+     in free-form mode (each episode generates its own subtasks +
+     memory),
+  4. uploads the annotated dataset to ``--dest_repo_id`` (when set)
+     or back to ``--repo_id``.
+
+Usage:
+
+    HF_TOKEN=hf_... uv run python examples/annotations/run_hf_job.py
+
+Adjust ``CMD`` below to point at your own dataset / target hub repo.
+"""
+
+import os
+
+from huggingface_hub import get_token, run_job
+
+token = os.environ.get("HF_TOKEN") or get_token()
+if not token:
+    raise RuntimeError("No HF token. Run `huggingface-cli login` or `export HF_TOKEN=hf_...`")
+
+CMD = (
+    "apt-get update -qq && apt-get install -y -qq git ffmpeg && "
+    "pip install --no-deps "
+    "'lerobot @ git+https://github.com/huggingface/lerobot.git@feat/language-annotation-pipeline' && "
+    "pip install --upgrade-strategy only-if-needed "
+    "datasets pyarrow av jsonlines draccus gymnasium torchcodec mergedeep pyyaml-include toml typing-inspect "
+    "openai && "
+    "export VLLM_MEMORY_PROFILER_ESTIMATE_CUDAGRAPHS=0 && "
+    "export VLLM_VIDEO_BACKEND=pyav && "
+    "lerobot-annotate "
+    "--repo_id=pepijn223/robocasa_pretrain_human300_v4 "
+    "--dest_repo_id=pepijn223/robocasa_pretrain_human300_v4_annotated5 "
+    "--push_to_hub=true "
+    "--vlm.backend=openai "
+    "--vlm.model_id=Qwen/Qwen3.6-27B "
+    "--vlm.parallel_servers=4 "
+    "--vlm.num_gpus=4 "
+    '--vlm.serve_command="vllm serve Qwen/Qwen3.6-27B '
+    "--tensor-parallel-size 1 --max-model-len 32768 "
+    '--gpu-memory-utilization 0.8 --uvicorn-log-level warning --port {port}" '
+    "--vlm.serve_ready_timeout_s=1800 "
+    "--vlm.client_concurrency=128 "
+    "--vlm.max_new_tokens=512 "
+    "--vlm.temperature=0.7 "
+    "--executor.episode_parallelism=16 "
+    "--vlm.chat_template_kwargs='{\"enable_thinking\": false}' "
+    "--vlm.camera_key=observation.images.robot0_agentview_right "
+    # Phase 1 — plan module (subtasks + plan + memory).
+    # Embed decoded frames directly (use_video_url=false) rather than
+    # handing the server a file:// clip. The embedded path is more
+    # reliable: if clip extraction ever fails, the video_url path would
+    # silently send NO video and the VLM would hallucinate subtasks from
+    # the task text alone.
+    #
+    # CONTEXT BUDGET: with embedded frames, each frame is ~250-320 vision
+    # tokens. The model's context is 32768 (see --max-model-len). 32
+    # frames sampled uniformly across the episode (~8-10k tokens) fits
+    # comfortably alongside the prompt and the describe pass.
+    # Do NOT raise max_video_frames toward 128 with embedded frames — that
+    # is ~33-39k tokens and overflows the context (BadRequestError 400,
+    # "Input length exceeds maximum context length").
+    "--plan.use_video_url=false "
+    "--plan.frames_per_second=1.0 "
+    "--plan.max_video_frames=32 "
+    # Constant 1 fps density via windowing: episodes longer than 32s are
+    # split into 32-second windows (each 32 frames @ 1 fps, fits context),
+    # so long episodes get MORE subtasks instead of a sparser whole-episode
+    # view. describe->segment runs per window; spans are merged +
+    # stitched to a contiguous whole-episode cover. 0 disables.
+    "--plan.subtask_window_seconds=32 "
+    # IMPORTANT for RoboCasa: the dataset's task string ("Navigate to the
+    # stove", "Pick the mug...") is authoritative and is what eval uses.
+    # ``derive_task_from_video=off`` keeps that canonical task driving
+    # subtask generation. Do NOT use ``always`` here — it throws the real
+    # task away, asks the VLM "what is this video about?" with no hint,
+    # and the hallucinated task then poisons every subtask + plan row.
+    "--plan.derive_task_from_video=off "
+    # NO task augmentation for RoboCasa: eval conditions on the exact task
+    # strings, so synthetic rephrasings are unused at best and (when they
+    # drift, e.g. "wander around the kitchen") harmful. 0 rephrasings +
+    # axes disabled = the policy only ever sees the canonical task.
+    "--plan.n_task_rephrasings=0 "
+    # action_records OFF: the structured {verb,object,arm,grasp,dest}
+    # schema is a manipulation schema; RoboCasa navigation / atomic tasks
+    # don't fit it and the VLM hallucinates. When on, records are purely
+    # additive (emitted as style="action_record" rows) and never touch
+    # the subtask text — useful only for long composite manipulation
+    # tasks. Leave off for RoboCasa atomic / navigation.
+    # Keep subtask decomposition tight for atomic tasks:
+    "--plan.plan_max_steps=10 "
+    # Only annotate subtasks + memory — skip the numbered "plan" rows
+    # (and their per-boundary VLM call). Flip to true to re-enable plan.
+    "--plan.emit_plan=false "
+    # NOTE: the grounding pass (describe -> segment, +1 VLM call/episode)
+    # is ON BY DEFAULT. Pass --plan.subtask_describe_first=false to disable
+    # on datasets you've verified are easy and want fewer calls.
+    # Phase 2 — interjections + speech.
+    "--interjections.max_interjections_per_episode=6 "
+    # Phase 4 — general VQA: DISABLED for this run.
+    "--vqa.enabled=false"
+)
+
+job = run_job(
+    image="vllm/vllm-openai:latest",
+    command=["bash", "-c", CMD],
+    flavor="h200x4",
+    secrets={"HF_TOKEN": token},
+    timeout="2h",
+)
+print(f"Job URL: {job.url}")
+print(f"Job ID:  {job.id}")

examples/dataset/create_progress_videos.py

@@ -15,10 +15,12 @@
 # limitations under the License.
 
 """
-Create MP4 (or GIF) videos with sarm_progress overlay for specified episodes.
+Create MP4 (or GIF) videos with per-frame progress overlay for specified episodes.
 
 Downloads datasets from HuggingFace, seeks directly into the episode segment
 of the source video, draws a progress line on each frame, and writes the result.
+The progress data is read from a parquet file that lives alongside the dataset
+(configurable via ``--progress-file``).
 
 Usage:
     python examples/dataset/create_progress_videos.py \
@@ -56,22 +58,26 @@ SCORE_FONT_SCALE = 0.8
 TASK_FONT_SCALE = 0.55
 
 
-def download_episode_metadata(repo_id: str, episode: int) -> Path:
-    """Download only the metadata and sarm_progress files for a dataset.
+def download_episode_metadata(
+    repo_id: str, episode: int, progress_file: str = "sarm_progress.parquet"
+) -> Path:
+    """Download only the metadata and per-frame progress file for a dataset.
 
     Args:
         repo_id: HuggingFace dataset repository ID.
         episode: Episode index (used for logging only; all meta is fetched).
+        progress_file: Filename of the per-frame progress parquet inside the
+            dataset repo.
 
     Returns:
         Local cache path for the downloaded snapshot.
     """
-    logging.info("[1/4] Downloading metadata for %s (episode %d) ...", repo_id, episode)
+    logging.info("[1/4] Downloading metadata + %s for %s (episode %d) ...", progress_file, repo_id, episode)
     local_path = Path(
         snapshot_download(
             repo_id=repo_id,
             repo_type="dataset",
-            allow_patterns=["meta/**", "sarm_progress.parquet"],
+            allow_patterns=["meta/**", progress_file],
             ignore_patterns=["*.mp4"],
         )
     )
@@ -215,25 +221,28 @@ def download_video_file(repo_id: str, local_path: Path, video_rel: str) -> Path:
     return video_path
 
 
-def load_progress_data(local_path: Path, episode: int) -> np.ndarray | None:
-    """Load sarm_progress values for an episode.
+def load_progress_data(
+    local_path: Path, episode: int, progress_file: str = "sarm_progress.parquet"
+) -> np.ndarray | None:
+    """Load per-frame progress values for an episode.
 
     Args:
         local_path: Dataset cache root.
         episode: Episode index.
+        progress_file: Filename of the per-frame progress parquet.
 
     Returns:
         Sorted (N, 2) array of (frame_index, progress), or None if unavailable.
     """
-    parquet_path = local_path / "sarm_progress.parquet"
+    parquet_path = local_path / progress_file
     if not parquet_path.exists():
-        logging.warning("sarm_progress.parquet not found")
+        logging.warning("%s not found", progress_file)
         return None
     df = pd.read_parquet(parquet_path)
-    logging.info("   sarm_progress.parquet columns: %s", list(df.columns))
+    logging.info("   %s columns: %s", progress_file, list(df.columns))
     episode_df = df[df["episode_index"] == episode].copy()
     if episode_df.empty:
-        logging.warning("No sarm_progress rows for episode %d", episode)
+        logging.warning("No progress rows for episode %d in %s", episode, progress_file)
         return None
     episode_df = episode_df.sort_values("frame_index")
 
@@ -576,6 +585,7 @@ def process_dataset(
     camera_key: str | None,
     output_dir: Path,
     create_gif: bool = False,
+    progress_file: str = "sarm_progress.parquet",
 ) -> Path | None:
     """Full pipeline: download, extract metadata, composite progress, write output.
 
@@ -585,6 +595,8 @@ def process_dataset(
         camera_key: Camera key to use, or None for auto-selection.
         output_dir: Directory to write output files.
         create_gif: If True, also generate a GIF from the MP4.
+        progress_file: Filename of the per-frame progress parquet inside the
+            dataset repo.
 
     Returns:
         Path to the final output file, or None on failure.
@@ -592,7 +604,7 @@ def process_dataset(
     safe_name = repo_id.replace("/", "_")
     logging.info("Processing: %s  |  episode %d", repo_id, episode)
 
-    local_path = download_episode_metadata(repo_id, episode)
+    local_path = download_episode_metadata(repo_id, episode, progress_file)
     logging.info("   Local cache: %s", local_path)
 
     episode_meta = load_episode_meta(local_path, episode, camera_key)
@@ -600,9 +612,9 @@ def process_dataset(
 
     video_path = download_video_file(repo_id, local_path, episode_meta["video_rel"])
 
-    progress_data = load_progress_data(local_path, episode)
+    progress_data = load_progress_data(local_path, episode, progress_file)
     if progress_data is None:
-        logging.error("Could not load sarm_progress data. Skipping overlay.")
+        logging.error("Could not load progress data from %s. Skipping overlay.", progress_file)
         return None
 
     logging.info("   Progress frames: %d", len(progress_data))
@@ -627,7 +639,7 @@ def process_dataset(
 
 def main() -> None:
     parser = argparse.ArgumentParser(
-        description="Create MP4/GIF videos with sarm_progress overlay for dataset episodes."
+        description="Create MP4/GIF videos with per-frame progress overlay for dataset episodes."
     )
     parser.add_argument(
         "--repo-id",
@@ -658,6 +670,15 @@ def main() -> None:
         action="store_true",
         help="Also generate a GIF from the MP4 output.",
     )
+    parser.add_argument(
+        "--progress-file",
+        type=str,
+        default="sarm_progress.parquet",
+        help=(
+            "Filename of the per-frame progress parquet inside the dataset repo "
+            "(default: 'sarm_progress.parquet')."
+        ),
+    )
     args = parser.parse_args()
 
     logging.basicConfig(level=logging.INFO, format="%(levelname)s: %(message)s")
@@ -670,6 +691,7 @@ def main() -> None:
         camera_key=args.camera_key,
         output_dir=args.output_dir,
         create_gif=args.gif,
+        progress_file=args.progress_file,
     )
 
     if result:

examples/dataset/slurm_compute_rabc.py

@@ -69,7 +69,7 @@ class ComputeProgressShards(PipelineStep):
         import torch
         from tqdm import tqdm
 
-        from lerobot.policies.sarm.compute_rabc_weights import (
+        from lerobot.rewards.sarm.compute_rabc_weights import (
             generate_all_frame_indices,
             interpolate_progress,
             load_sarm_resources,

examples/notebooks/quickstart.ipynb

@@ -80,7 +80,7 @@
     "}\n",
     "\n",
     "# Dataset\n",
-    "HF_USER = \"your_hf_username\"  # `huggingface-cli whoami` to find your username\n",
+    "HF_USER = \"your_hf_username\"  # `hf auth whoami` to find your username\n",
     "DATASET_NAME = \"my_so101_dataset\"\n",
     "TASK_DESCRIPTION = \"pick and place the block\"\n",
     "NUM_EPISODES = 10\n",
@@ -291,7 +291,34 @@
     "\n",
     "Uses `POLICY_PATH` from the Configuration cell (defaults to the Hub repo ID). You can also put there the `LAST_CHECKPOINT_PATH`.\n",
     "\n",
-    "See the [inference docs](https://huggingface.co/docs/lerobot/il_robots#run-inference-and-evaluate-your-policy) for details."
+    "See the [inference docs](https://huggingface.co/docs/lerobot/il_robots#run-inference-and-evaluate-your-policy) for details.\n",
+    "\n",
+    "Recently ```lerobot-rollout``` was introduced, you can [read more about it here](https://huggingface.co/docs/lerobot/main/en/il_robots?eval=Base+mode+%28no+recording%29#run-inference-and-evaluate-your-policy)."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "print_cmd(\n",
+    "    \"lerobot-rollout\",\n",
+    "    \"--strategy.type=base\",\n",
+    "    f\"--policy.path={POLICY_PATH}\",\n",
+    "    f\"--robot.type={ROBOT_TYPE}\",\n",
+    "    f\"--robot.port={ROBOT_PORT}\",\n",
+    "    CAMERAS_FLAG,\n",
+    "    f'--task=\"{TASK_DESCRIPTION}\"',\n",
+    "    \"--duration=60\",\n",
+    ")"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "if you are using the V0.5.1 release you should use ```lerobot-record``` instead of rollout"
    ]
   },
   {

examples/omx/README.md

@@ -0,0 +1,136 @@
+# OMX Follower — Cube Pick And Place Example
+
+This is an example of what is possible to do with LeRobot on a physical setup.
+It is a WIP and being used internally at LeRobot and specific to our setup, but we hope it can be a useful reference for how to use LeRobot APIs and CLIs.
+
+It includes an end-to-end example for the **OMX Follower** robot arm: pick and place a cube dataset, train a policy, and deploy it autonomously.
+
+## Hardware
+
+| Component | Value                                |
+| --------- | ------------------------------------ |
+| Robot     | OMX Follower                         |
+| Cameras   | 2× OpenCV cameras (wrist + top-down) |
+
+## Scripts
+
+| Script                 | Purpose                                                         |
+| ---------------------- | --------------------------------------------------------------- |
+| `reset_environment.py` | Standalone utility: sweep workspace, grab cube, place cube      |
+| `record_grab.py`       | Automated data collection: reset → place → record grab episodes |
+
+## Setup
+
+Make sure you have LeRobot installed in your env. (See [the installation guide](https://huggingface.co/docs/lerobot/installation))
+
+Next, we will declare some environment variables for convenience. Adjust the camera indices and robot port to match your system configuration.
+
+```bash
+export ROBOT_PORT=/dev/ttyACM0
+export TELEOP_PORT=/dev/ttyACM1
+export HF_USERNAME=<your_hf_username>
+export ROBOT_CAMERAS="{ wrist: {type: opencv, index_or_path: 0, width: 640, height: 480, fps: 30, fourcc: MJPG}, top: {type: opencv, index_or_path: 2, width: 640, height: 480, fps: 30, fourcc: MJPG} }"
+```
+
+## Step 1 — Collect Data
+
+```bash
+lerobot-record \
+    --robot.type=omx_follower \
+    --robot.port=$ROBOT_PORT \
+    --robot.id=omx_follower \
+    --robot.cameras="$ROBOT_CAMERAS" \
+    --teleop.type=omx_leader \
+    --teleop.port=$TELEOP_PORT \
+    --teleop.id=omx_leader \
+    --dataset.repo_id=$HF_USERNAME/omx_pickandplace \
+    --dataset.root=data/omx_pickandplace \
+    --dataset.num_episodes=50 \
+    --dataset.single_task="Pick the cube and place it in the blue square" \
+    --dataset.streaming_encoding=true \
+    --dataset.push_to_hub=true
+```
+
+### Bonus Auto-Collect script
+
+/!\ This is specific to our setup and the task of picking and placing a cube. It is not a general-purpose data collection script. As you may notice, it doesn't require a teleop.
+
+```bash
+python -m examples.omx.record_grab \
+    --robot.type=omx_follower \
+    --robot.port=$ROBOT_PORT \
+    --robot.id=omx_follower \
+    --robot.cameras="$ROBOT_CAMERAS" \
+    --dataset.repo_id=$HF_USERNAME/omx_pickandplace \
+    --dataset.root=data/omx_pickandplace \
+    --dataset.num_episodes=50 \
+    --dataset.single_task="Pick the cube and place it in the blue square" \
+    --dataset.streaming_encoding=true \
+    --dataset.push_to_hub=true
+```
+
+Each episode:
+
+1. The arm grabs the cube from the center of the workspace and places it at a random position.
+2. The arm returns to HOME.
+3. A targeted grab is recorded: HOME → approach raised → lower onto cube → grasp → lift → carry → drop → HOME.
+
+A dataset is already available here [`maximellerbach/omx_pickandplace`](https://huggingface.co/datasets/maximellerbach/omx_pickandplace), so you can skip directly to training if you want.
+
+## Step 2 — Train
+
+To train a simple `ACT` policy on the collected dataset, you can use the `lerobot-train` CLI:
+
+```bash
+lerobot-train \
+    --dataset.repo_id=$HF_USERNAME/omx_pickandplace \
+    --policy.type=act \
+    --output_dir=outputs/train/omx_pickandplace_act \
+    --policy.device=cuda \
+    --policy.repo_id=$HF_USERNAME/omx_pickandplace_act \
+    --steps=20000 \
+    --wandb.enable=true
+```
+
+A pretrained `ACT` policy is already available here [`maximellerbach/omx_pickandplace_act`](https://huggingface.co/maximellerbach/omx_pickandplace_act).
+
+## Step 3 — Rollout
+
+Use the `lerobot-rollout` CLI with base strategy:
+
+```bash
+lerobot-rollout \
+    --strategy.type=base \
+    --robot.type=omx_follower \
+    --robot.port=$ROBOT_PORT \
+    --robot.id=omx_follower \
+    --robot.cameras="$ROBOT_CAMERAS" \
+    --policy.path=$HF_USERNAME/omx_pickandplace_act \
+```
+
+For continuous recording with automatic upload (sentry mode):
+
+```bash
+lerobot-rollout \
+    --strategy.type=sentry \
+    --strategy.upload_every_n_episodes=10 \
+    --robot.type=omx_follower \
+    --robot.port=$ROBOT_PORT \
+    --robot.id=omx_follower \
+    --robot.cameras="$ROBOT_CAMERAS" \
+    --policy.path=$HF_USERNAME/omx_pickandplace_act \
+    --dataset.repo_id=$HF_USERNAME/rollout_omx_pickandplace_act \
+```
+
+## Environment Reset Utility
+
+Those are specific to this particular physical setup. Those are scripts that execute hardcoded sequences of actions on the robot to reset the environment, which is useful for data collection and evaluation. They are not general-purpose scripts.
+
+`reset_environment.py` can be run standalone to prepare the workspace:
+
+```bash
+# Grab cube + place it at a random position on the left side
+python -m examples.omx.reset_environment --port $ROBOT_PORT --mode grab_and_place
+```
+
+It also exposes `grab_cube(robot)` and `place_cube(robot)` for use in custom scripts.

examples/omx/record_grab.py

@@ -0,0 +1,422 @@
+#!/usr/bin/env python3
+"""
+Auto-record grab episodes for the OMX robot arm.
+
+Each episode cycle:
+  1. grab_and_place  — grab cube from workspace center and place at a random (pan, reach) position
+  2. HOME            — return arm to home with gripper open
+  3. record_grab     — execute a targeted grab to the stored position while recording
+                       observations + actions to a LeRobotDataset
+
+Usage (run from repo root):
+    python -m examples.omx.record_grab \\
+        --robot.type=omx_follower \\
+        --robot.port=/dev/ttyACM0 \\
+        --robot.id=omx_follower \\
+        --robot.cameras="{ wrist: {type: opencv, index_or_path: 6, width: 640, height: 480, fps: 30, fourcc: MJPG}, top: {type: opencv, index_or_path: 4, width: 640, height: 480, fps: 30, fourcc: MJPG} }" \\
+        --dataset.repo_id=<hf_username>/<dataset_name> \\
+        --dataset.root=data/omx_grab \\
+        --dataset.num_episodes=50 \\
+        --dataset.single_task="Grab the cube" \\
+        --dataset.streaming_encoding=true
+"""
+
+import logging
+from dataclasses import dataclass
+from pprint import pformat
+
+import numpy as np
+
+from lerobot.cameras import CameraConfig  # noqa: F401
+from lerobot.cameras.opencv import OpenCVCameraConfig  # noqa: F401
+from lerobot.configs import parser
+from lerobot.configs.dataset import DatasetRecordConfig
+from lerobot.datasets import (
+    LeRobotDataset,
+    VideoEncodingManager,
+    aggregate_pipeline_dataset_features,
+    create_initial_features,
+)
+from lerobot.processor import make_default_processors
+from lerobot.robots import RobotConfig, make_robot_from_config
+from lerobot.robots.omx_follower import OmxFollower
+from lerobot.utils.constants import ACTION, OBS_STR
+from lerobot.utils.feature_utils import build_dataset_frame, combine_feature_dicts
+from lerobot.utils.robot_utils import precise_sleep
+
+from .reset_environment import (
+    APPROACH_SPEED,
+    GRIPPER_CLOSE_POS,
+    HOME_POSE,
+    PUSH_END_ELBOW_FLEX,
+    PUSH_END_SHOULDER_LIFT,
+    PUSH_START_ELBOW_FLEX,
+    PUSH_START_SHOULDER_LIFT,
+    array_to_pose,
+    grab_cube,
+    horizontal_wrist_flex,
+    move_to_pose,
+    place_cube,
+    pose_to_array,
+)
+
+# ── Grab-episode motion parameters ────────────────────────────────────────────
+
+# Shoulder-lift offset for the raised approach phase (subtracted from the target sl, arm is higher).
+GRAB_RAISE_SL_OFFSET = 20.0
+GRAB_LOWER_SPEED = 20.0
+RECORD_SPEED = 30.0
+
+# Pose the arm travels to after closing the gripper (cube held).
+GRAB_CARRY_POSE = {
+    "shoulder_pan.pos": -23.0,
+    "shoulder_lift.pos": 5.0,
+    "elbow_flex.pos": 18.0,
+    "wrist_flex.pos": -14.0,
+    "wrist_roll.pos": 0.0,
+    "gripper.pos": GRIPPER_CLOSE_POS,
+}
+
+# Per-joint jitter limits (degrees) applied to transit waypoints for human-like variation.
+# Cube-approach and carry poses are never jittered to preserve precision.
+_JITTER_LIMITS: dict[str, float] = {
+    "shoulder_pan.pos": 5.0,
+    "shoulder_lift.pos": 4.0,
+    "elbow_flex.pos": 4.0,
+    "wrist_flex.pos": 3.0,
+    "wrist_roll.pos": 2.0,
+    "gripper.pos": 0.0,
+}
+
+
+def _jitter_pose(pose: dict, rng: np.random.Generator) -> dict:
+    """Return a copy of pose with independent per-joint random perturbations."""
+    return {
+        k: v + rng.uniform(-_JITTER_LIMITS.get(k, 0.0), _JITTER_LIMITS.get(k, 0.0)) for k, v in pose.items()
+    }
+
+
+def _random_stuck_pose(rng: np.random.Generator) -> dict:
+    """Return a physically plausible stuck pose (failed grasp), gripper closed.
+
+    ef bounds are piecewise-linear in sl so the arm stays in a reachable,
+    table-safe envelope across the full sl range:
+      sl=-50 → ef ∈ [  0,  50]   (arm raised, can be bent forward)
+      sl=  0 → ef ∈ [-25,  25]   (mid reach)
+      sl= 30 → ef ∈ [-20,   0]   (arm extended, little room to flex)
+    wrist_flex is randomly offset from the horizontal value.
+    """
+    pan = float(rng.uniform(-5.0, 35.0))
+    sl = float(rng.uniform(-50.0, 30.0))
+
+    if sl <= 0.0:
+        alpha = (sl + 50.0) / 50.0  # 0 at sl=-50, 1 at sl=0
+        ef_lo = alpha * -25.0  # 0 → -25
+        ef_hi = 50.0 + alpha * -25.0  # 50 → 25
+    else:
+        alpha = sl / 30.0  # 0 at sl=0, 1 at sl=30
+        ef_lo = -25.0 + alpha * 5.0  # -25 → -20
+        ef_hi = 25.0 + alpha * -25.0  # 25 → 0
+
+    ef = float(rng.uniform(ef_lo, ef_hi))
+    wf = horizontal_wrist_flex(sl, ef) + float(rng.uniform(-15.0, 15.0))
+    return {
+        "shoulder_pan.pos": pan,
+        "shoulder_lift.pos": sl,
+        "elbow_flex.pos": ef,
+        "wrist_flex.pos": wf,
+        "wrist_roll.pos": float(rng.uniform(-15.0, 15.0)),
+        "gripper.pos": GRIPPER_CLOSE_POS,
+    }
+
+
+logger = logging.getLogger(__name__)
+
+
+@dataclass
+class OmxRecordGrabConfig:
+    robot: RobotConfig
+    dataset: DatasetRecordConfig
+    # Resume recording on an existing dataset.
+    resume: bool = False
+    # Fraction of episodes that start from a random stuck pose (gripper closed) to
+    # generate recovery data.  0.0 = disabled, 1.0 = all episodes are recovery starts.
+    recovery_prob: float = 0.5
+
+
+def record_episode_spline(
+    robot: OmxFollower,
+    waypoints: list[dict],
+    speeds: list[float],
+    dataset: LeRobotDataset,
+    task: str,
+) -> None:
+    """Execute a Catmull-Rom-style spline through waypoints, recording each frame.
+
+    Segment durations are parameterized from the maximum absolute joint delta
+    between consecutive waypoints divided by the requested segment speed,
+    producing non-uniform timing in joint space. Interior tangents are derived
+    from the adjacent per-segment velocities, with clamped (zero-velocity)
+    endpoints so the arm starts and stops smoothly. Each segment is cubic
+    Hermite, giving C1 continuity at every waypoint.
+    """
+    pts = [pose_to_array(w) for w in waypoints]
+    n = len(pts)
+
+    # Steps and duration per segment
+    n_steps_list = []
+    timestamps = []
+    for i in range(n - 1):
+        max_dist = float(np.max(np.abs(pts[i + 1] - pts[i])))
+        ns = max(1, int(max_dist / speeds[i] * dataset.fps)) if max_dist >= 0.5 else 0
+        n_steps_list.append(ns)
+        timestamps.append(ns / dataset.fps)
+
+    # Velocity tangents (deg/sec) — clamped at endpoints, Catmull-Rom for interior
+    vels = [np.zeros_like(pts[0])]
+    for i in range(1, n - 1):
+        v_prev = (pts[i] - pts[i - 1]) / timestamps[i - 1] if timestamps[i - 1] > 0 else np.zeros_like(pts[0])
+        v_next = (pts[i + 1] - pts[i]) / timestamps[i] if timestamps[i] > 0 else np.zeros_like(pts[0])
+        vels.append(0.5 * (v_prev + v_next))
+    vels.append(np.zeros_like(pts[0]))
+
+    dt = 1.0 / dataset.fps
+    for seg in range(n - 1):
+        ns = n_steps_list[seg]
+        if ns == 0:
+            continue
+        p0, p1 = pts[seg], pts[seg + 1]
+        # Scale velocity (deg/sec) to t-space tangent (deg/t-unit, where t: 0→1 over ns steps)
+        m0 = vels[seg] * timestamps[seg]
+        m1 = vels[seg + 1] * timestamps[seg]
+
+        for step in range(1, ns + 1):
+            t = step / ns
+            h00 = 2 * t**3 - 3 * t**2 + 1
+            h10 = t**3 - 2 * t**2 + t
+            h01 = -2 * t**3 + 3 * t**2
+            h11 = t**3 - t**2
+            commanded = h00 * p0 + h10 * m0 + h01 * p1 + h11 * m1
+
+            action = array_to_pose(commanded)
+            robot.send_action(action)
+            obs = robot.get_observation()
+            obs_frame = build_dataset_frame(dataset.features, obs, prefix=OBS_STR)
+            action_frame = build_dataset_frame(dataset.features, action, prefix=ACTION)
+            dataset.add_frame({**obs_frame, **action_frame, "task": task})
+            precise_sleep(dt)
+
+
+def record_grab_episode(
+    robot: OmxFollower,
+    dataset: LeRobotDataset,
+    pan: float,
+    t: float,
+    task: str,
+    recovery_start: bool = False,
+) -> None:
+    """Execute a targeted grab to the stored (pan, t) position, recording every frame.
+
+    Normal sequence (initial HOME move is NOT recorded):
+      HOME → raised approach above cube → lower → close gripper
+           → raise [jittered] → retract [jittered] → GRAB_CARRY_POSE → drop → HOME
+
+    Recovery sequence (recovery_start=True): arm is moved to a random stuck pose
+    (gripper closed) without recording, then recording begins from there:
+      stuck_pose → raised approach above cube → [normal grab sequence from there]
+
+    All segments are joined by a Catmull-Rom spline (C1-continuous velocities).
+    """
+    sl = PUSH_START_SHOULDER_LIFT + t * (PUSH_END_SHOULDER_LIFT - PUSH_START_SHOULDER_LIFT)
+    ef = PUSH_START_ELBOW_FLEX + t * (PUSH_END_ELBOW_FLEX - PUSH_START_ELBOW_FLEX)
+    sl_raised = sl - GRAB_RAISE_SL_OFFSET
+    wf_horizontal = horizontal_wrist_flex(sl, ef)
+
+    rng = np.random.default_rng()
+
+    if recovery_start:
+        stuck_pose = _random_stuck_pose(rng)
+        logger.info(f"Recovery start: {stuck_pose}")
+        move_to_pose(robot, stuck_pose, APPROACH_SPEED)
+        first_waypoints = [stuck_pose]
+        first_speeds = []
+    else:
+        jittery_start = _jitter_pose(HOME_POSE, rng)
+        move_to_pose(robot, jittery_start, APPROACH_SPEED)
+        first_waypoints = [jittery_start]
+        first_speeds = []
+
+    waypoints = first_waypoints + [
+        {  # raised approach: arm above cube
+            "shoulder_pan.pos": pan,
+            "shoulder_lift.pos": sl_raised,
+            "elbow_flex.pos": ef,
+            "wrist_flex.pos": horizontal_wrist_flex(sl_raised, ef),
+            "wrist_roll.pos": 0.0,
+            "gripper.pos": 60.0,
+        },
+        {  # lower onto cube — no jitter: precision needed
+            "shoulder_pan.pos": pan,
+            "shoulder_lift.pos": sl,
+            "elbow_flex.pos": ef,
+            "wrist_flex.pos": wf_horizontal,
+            "wrist_roll.pos": 0.0,
+            "gripper.pos": 60.0,
+        },
+        {  # close gripper — no jitter: precision needed
+            "shoulder_pan.pos": pan,
+            "shoulder_lift.pos": sl,
+            "elbow_flex.pos": ef,
+            "wrist_flex.pos": wf_horizontal,
+            "wrist_roll.pos": 0.0,
+            "gripper.pos": GRIPPER_CLOSE_POS,
+        },
+        _jitter_pose(
+            {  # raise with cube
+                "shoulder_pan.pos": pan,
+                "shoulder_lift.pos": sl_raised,
+                "elbow_flex.pos": ef,
+                "wrist_flex.pos": horizontal_wrist_flex(sl_raised, ef),
+                "wrist_roll.pos": 0.0,
+                "gripper.pos": GRIPPER_CLOSE_POS,
+            },
+            rng,
+        ),
+        _jitter_pose(
+            {  # retract: fold arm toward HOME before sweeping to carry zone
+                "shoulder_pan.pos": pan * 0.25,
+                "shoulder_lift.pos": HOME_POSE["shoulder_lift.pos"] + 5.0,
+                "elbow_flex.pos": HOME_POSE["elbow_flex.pos"] - 5.0,
+                "wrist_flex.pos": 0.0,
+                "wrist_roll.pos": 0.0,
+                "gripper.pos": GRIPPER_CLOSE_POS,
+            },
+            rng,
+        ),
+        GRAB_CARRY_POSE,  # no jitter: target drop zone
+        {**GRAB_CARRY_POSE, "gripper.pos": 60.0},  # drop cube
+        HOME_POSE,
+    ]
+    speeds = first_speeds + [
+        RECORD_SPEED,  # (HOME →) raised approach
+        GRAB_LOWER_SPEED,  # raised approach → lower
+        GRAB_LOWER_SPEED,  # lower → close gripper
+        RECORD_SPEED,  # close gripper → raise
+        RECORD_SPEED,  # raise → retract
+        RECORD_SPEED,  # retract → carry pose
+        RECORD_SPEED,  # carry pose → drop
+        RECORD_SPEED,  # drop → HOME
+    ]
+
+    record_episode_spline(robot, waypoints, speeds, dataset, task)
+
+    # Dwell at HOME for ~0.5 s before next episode
+    home_action = build_dataset_frame(dataset.features, HOME_POSE, prefix=ACTION)
+    dt = 1.0 / dataset.fps
+    for _ in range(int(dataset.fps * 0.5)):
+        robot.send_action(HOME_POSE)
+        obs = robot.get_observation()
+        obs_frame = build_dataset_frame(dataset.features, obs, prefix=OBS_STR)
+        dataset.add_frame({**obs_frame, **home_action, "task": task})
+        precise_sleep(dt)
+
+
+@parser.wrap()
+def record_grab(cfg: OmxRecordGrabConfig) -> LeRobotDataset:
+    logging.basicConfig(level=logging.INFO, format="%(levelname)s: %(message)s")
+    logger.info(pformat(cfg))
+
+    robot = make_robot_from_config(cfg.robot)
+    use_videos = cfg.dataset.video
+
+    teleop_action_processor, _, robot_obs_processor = make_default_processors()
+
+    dataset_features = combine_feature_dicts(
+        aggregate_pipeline_dataset_features(
+            pipeline=teleop_action_processor,
+            initial_features=create_initial_features(action=robot.action_features),
+            use_videos=use_videos,
+        ),
+        aggregate_pipeline_dataset_features(
+            pipeline=robot_obs_processor,
+            initial_features=create_initial_features(observation=robot.observation_features),
+            use_videos=use_videos,
+        ),
+    )
+
+    num_cameras = len(robot.cameras) if hasattr(robot, "cameras") else 0
+    dataset = None
+
+    try:
+        if cfg.resume:
+            dataset = LeRobotDataset.resume(
+                cfg.dataset.repo_id,
+                root=cfg.dataset.root,
+                streaming_encoding=cfg.dataset.streaming_encoding,
+                batch_encoding_size=cfg.dataset.video_encoding_batch_size,
+                vcodec=cfg.dataset.vcodec,
+                encoder_threads=cfg.dataset.encoder_threads,
+                image_writer_processes=cfg.dataset.num_image_writer_processes if num_cameras > 0 else 0,
+                image_writer_threads=cfg.dataset.num_image_writer_threads_per_camera * num_cameras
+                if num_cameras > 0
+                else 0,
+            )
+        else:
+            cfg.dataset.stamp_repo_id()
+            dataset = LeRobotDataset.create(
+                cfg.dataset.repo_id,
+                cfg.dataset.fps,
+                root=cfg.dataset.root,
+                robot_type=robot.name,
+                features=dataset_features,
+                use_videos=use_videos,
+                streaming_encoding=cfg.dataset.streaming_encoding,
+                batch_encoding_size=cfg.dataset.video_encoding_batch_size,
+                vcodec=cfg.dataset.vcodec,
+                encoder_threads=cfg.dataset.encoder_threads,
+                image_writer_processes=cfg.dataset.num_image_writer_processes if num_cameras > 0 else 0,
+                image_writer_threads=cfg.dataset.num_image_writer_threads_per_camera * num_cameras
+                if num_cameras > 0
+                else 0,
+            )
+
+        robot.connect(calibrate=True)
+
+        rng = np.random.default_rng()
+        with VideoEncodingManager(dataset):
+            for episode_idx in range(cfg.dataset.num_episodes):
+                logger.info(f"=== Episode {episode_idx + 1}/{cfg.dataset.num_episodes} ===")
+
+                logger.info("Step 1: grabbing and placing cube...")
+                grab_cube(robot)
+                pan, t = place_cube(robot)
+                logger.info(f"Cube placed at pan={pan:.1f}, reach={t:.2f}")
+
+                recovery_start = cfg.recovery_prob > 0 and float(rng.random()) < cfg.recovery_prob
+                logger.info(f"Step 2: recording {'recovery ' if recovery_start else ''}grab episode...")
+                record_grab_episode(
+                    robot,
+                    dataset,
+                    pan,
+                    t,
+                    cfg.dataset.single_task,
+                    recovery_start=recovery_start,
+                )
+
+                dataset.save_episode()
+                logger.info(f"Episode {episode_idx + 1} saved.")
+
+    finally:
+        if dataset:
+            dataset.finalize()
+        if robot.is_connected:
+            robot.disconnect()
+
+    if cfg.dataset.push_to_hub and dataset and dataset.num_episodes > 0:
+        dataset.push_to_hub(tags=cfg.dataset.tags, private=cfg.dataset.private)
+
+    return dataset
+
+
+if __name__ == "__main__":
+    record_grab()

examples/omx/reset_environment.py

@@ -0,0 +1,267 @@
+#!/usr/bin/env python3
+"""
+Auto-reset and cube-grab utility for the OMX robot arm.
+
+Provides:
+  - grab_cube(robot): sweep workspace, center cube, close gripper
+  - place_cube(robot): carry cube to a random position, release
+
+Standalone usage (run from repo root):
+    python -m examples.omx.reset_environment --port /dev/ttyACM1 --mode grab
+    python -m examples.omx.reset_environment --port /dev/ttyACM1 --mode grab_and_place
+
+Joint range: -100 to 100 for arm joints; gripper: 50 = closed, 80 = open.
+
+To read current joint values for calibration, add after robot.connect():
+    obs = robot.get_observation()
+    print({k: round(obs[k], 1) for k in JOINT_NAMES})
+    robot.disconnect(); raise SystemExit
+
+Parallel-to-ground IK: wrist_flex = WRIST_HORIZONTAL_OFFSET - shoulder_lift - elbow_flex.
+Linear interpolation preserves this constraint between any two poses that satisfy it.
+"""
+
+import argparse
+import logging
+
+import numpy as np
+
+from lerobot.robots.omx_follower import OmxFollower, OmxFollowerConfig
+from lerobot.robots.robot import Robot
+from lerobot.utils.robot_utils import precise_sleep
+
+logger = logging.getLogger(__name__)
+
+# ── Poses ─────────────────────────────────────────────────────────────────────
+
+HOME_POSE = {
+    "shoulder_pan.pos": 0.0,
+    "shoulder_lift.pos": -50.0,
+    "elbow_flex.pos": 50.0,
+    "wrist_flex.pos": 0.0,
+    "wrist_roll.pos": 0.0,
+    "gripper.pos": 60.0,
+}
+
+SWEEP_WAYPOINTS = [
+    {
+        "shoulder_pan.pos": -60.0,
+        "shoulder_lift.pos": 50.0,
+        "elbow_flex.pos": -60.0,
+        "wrist_flex.pos": -20.0,
+        "wrist_roll.pos": 0.0,
+        "gripper.pos": 60.0,
+    },
+    {
+        "shoulder_pan.pos": -30.0,
+        "shoulder_lift.pos": 50.0,
+        "elbow_flex.pos": -60.0,
+        "wrist_flex.pos": -5.0,
+        "wrist_roll.pos": 0.0,
+        "gripper.pos": 60.0,
+    },
+    {
+        "shoulder_pan.pos": 20.0,
+        "shoulder_lift.pos": 50.0,
+        "elbow_flex.pos": -55.0,
+        "wrist_flex.pos": -5.0,
+        "wrist_roll.pos": 0.0,
+        "gripper.pos": 60.0,
+    },
+]
+
+# ── Motion parameters ─────────────────────────────────────────────────────────
+
+CONTROL_HZ = 30
+APPROACH_SPEED = 50.0
+SWEEP_SPEED = 40.0
+
+# ── Grab-sequence parameters ──────────────────────────────────────────────────
+
+GRAB_PAN = 0.0
+SWEEP_LEFT_PAN = -60.0
+SWEEP_RIGHT_PAN = 60.0
+SWEEP_END_OFFSET = 5.0  # stop before center so the cube isn't pushed past GRAB_PAN
+SWEEP_END_PAN_RANGE = (15.0, 20.0)
+
+SWEEP_LOW_SHOULDER_LIFT = 50.0
+SWEEP_LOW_ELBOW_FLEX_START = -60.0
+SWEEP_LOW_ELBOW_FLEX_END = -55.0
+
+SWEEP_HIGH_WRIST_FLEX = -20.0  # wrist tilted up during high approach to clear obstacles
+
+PUSH_START_SHOULDER_LIFT = 0.0
+PUSH_START_ELBOW_FLEX = 45.0
+PUSH_END_SHOULDER_LIFT = 50.0
+PUSH_END_ELBOW_FLEX = -50.0
+# Subtracted from shoulder_lift during the push sweep to clear the platform surface.
+# Does not affect the grab-target interpolation in record_grab.py.
+PUSH_RAISE_OFFSET = 5.0
+
+WRIST_HORIZONTAL_OFFSET = 0.0  # tune if gripper tilts during push: + tilts nose up, - down
+GRIPPER_CLOSE_POS = 50.0
+
+PLACE_LEFT_PAN_RANGE = (5.0, 30.0)  # random pan range for cube placement on the left side
+PLACE_REACH_RANGE = (0.1, 0.7)  # 0 = arm retracted (PUSH_START), 1 = fully extended (PUSH_END)
+
+JOINT_NAMES = [
+    "shoulder_pan.pos",
+    "shoulder_lift.pos",
+    "elbow_flex.pos",
+    "wrist_flex.pos",
+    "wrist_roll.pos",
+    "gripper.pos",
+]
+
+# ── Helpers ───────────────────────────────────────────────────────────────────
+
+
+def pose_to_array(pose: dict) -> np.ndarray:
+    return np.array([pose[k] for k in JOINT_NAMES])
+
+
+def array_to_pose(arr: np.ndarray) -> dict:
+    return {k: float(arr[i]) for i, k in enumerate(JOINT_NAMES)}
+
+
+def horizontal_wrist_flex(shoulder_lift: float, elbow_flex: float) -> float:
+    return WRIST_HORIZONTAL_OFFSET - shoulder_lift - elbow_flex
+
+
+def _low_sweep_pose(pan: float, elbow_flex: float, wrist_flex: float | None = None) -> dict:
+    sl = SWEEP_LOW_SHOULDER_LIFT
+    return {
+        "shoulder_pan.pos": pan,
+        "shoulder_lift.pos": sl,
+        "elbow_flex.pos": elbow_flex,
+        "wrist_flex.pos": horizontal_wrist_flex(sl, elbow_flex) if wrist_flex is None else wrist_flex,
+        "wrist_roll.pos": 0.0,
+        "gripper.pos": 60.0,
+    }
+
+
+def _high_sweep_pose(pan: float) -> dict:
+    return {**HOME_POSE, "shoulder_pan.pos": pan, "wrist_flex.pos": SWEEP_HIGH_WRIST_FLEX}
+
+
+def _push_pose(shoulder_lift: float, elbow_flex: float, pan: float = GRAB_PAN, gripper: float = 70.0) -> dict:
+    return {
+        "shoulder_pan.pos": pan,
+        "shoulder_lift.pos": shoulder_lift,
+        "elbow_flex.pos": elbow_flex,
+        "wrist_flex.pos": horizontal_wrist_flex(shoulder_lift, elbow_flex),
+        "wrist_roll.pos": 0.0,
+        "gripper.pos": gripper,
+    }
+
+
+def move_to_pose(robot: Robot, target: dict, speed: float) -> None:
+    """Interpolate from current position to target at the given speed (units/s)."""
+    obs = robot.get_observation()
+    current = np.array([obs[k] for k in JOINT_NAMES])
+    goal = pose_to_array(target)
+
+    max_distance = float(np.max(np.abs(goal - current)))
+    if max_distance < 0.5:
+        return
+
+    n_steps = max(1, int(max_distance / speed * CONTROL_HZ))
+    dt = 1.0 / CONTROL_HZ
+    for step in range(1, n_steps + 1):
+        t = step / n_steps
+        robot.send_action(array_to_pose(current + t * (goal - current)))
+        precise_sleep(dt)
+
+
+# ── Sequences ─────────────────────────────────────────────────────────────────
+
+
+def grab_cube(robot: Robot) -> None:
+    """Left sweep → right sweep → extend arm parallel to ground → close gripper."""
+    move_to_pose(robot, HOME_POSE, APPROACH_SPEED)
+
+    for pan, end_pan in [
+        (SWEEP_LEFT_PAN, GRAB_PAN - SWEEP_END_OFFSET),
+        (SWEEP_RIGHT_PAN, GRAB_PAN + SWEEP_END_OFFSET),
+    ]:
+        logger.info(f"Sweeping {'left' if pan < 0 else 'right'} → center...")
+        move_to_pose(robot, _high_sweep_pose(pan), APPROACH_SPEED)
+        move_to_pose(
+            robot, _low_sweep_pose(pan, SWEEP_LOW_ELBOW_FLEX_START, wrist_flex=-20.0), APPROACH_SPEED
+        )
+        move_to_pose(robot, _low_sweep_pose(end_pan, SWEEP_LOW_ELBOW_FLEX_END, wrist_flex=0.0), SWEEP_SPEED)
+        move_to_pose(robot, HOME_POSE, APPROACH_SPEED)
+
+    logger.info("Extending to push cube into gripper...")
+    move_to_pose(
+        robot,
+        _push_pose(PUSH_START_SHOULDER_LIFT - PUSH_RAISE_OFFSET, PUSH_START_ELBOW_FLEX),
+        APPROACH_SPEED,
+    )
+    move_to_pose(
+        robot,
+        _push_pose(PUSH_END_SHOULDER_LIFT - PUSH_RAISE_OFFSET, PUSH_END_ELBOW_FLEX),
+        SWEEP_SPEED,
+    )
+
+    logger.info("Closing gripper...")
+    move_to_pose(
+        robot,
+        _push_pose(PUSH_END_SHOULDER_LIFT, PUSH_END_ELBOW_FLEX, gripper=GRIPPER_CLOSE_POS),
+        APPROACH_SPEED,
+    )
+
+    logger.info("Grab complete.")
+
+
+def place_cube(robot: Robot) -> tuple[float, float]:
+    """Carry the cube (gripper closed) to a random position on the left side, then release.
+
+    Returns:
+        (pan, t): pan angle and reach scalar [0, 1] of the placement position.
+    """
+    pan = float(np.random.uniform(*PLACE_LEFT_PAN_RANGE))
+    t = float(np.random.uniform(*PLACE_REACH_RANGE))
+    sl = PUSH_START_SHOULDER_LIFT + t * (PUSH_END_SHOULDER_LIFT - PUSH_START_SHOULDER_LIFT)
+    ef = PUSH_START_ELBOW_FLEX + t * (PUSH_END_ELBOW_FLEX - PUSH_START_ELBOW_FLEX)
+    logger.info(f"Placing cube at pan={pan:.1f}, reach={t:.2f}...")
+
+    move_to_pose(robot, {**HOME_POSE, "gripper.pos": GRIPPER_CLOSE_POS}, APPROACH_SPEED)
+    move_to_pose(
+        robot, {**HOME_POSE, "shoulder_pan.pos": pan, "gripper.pos": GRIPPER_CLOSE_POS}, APPROACH_SPEED
+    )
+    move_to_pose(robot, _push_pose(sl, ef, pan=pan, gripper=GRIPPER_CLOSE_POS), APPROACH_SPEED)
+    move_to_pose(robot, _push_pose(sl, ef, pan=pan, gripper=80.0), APPROACH_SPEED)
+    move_to_pose(robot, HOME_POSE, APPROACH_SPEED)
+    logger.info("Place complete.")
+    return pan, t
+
+
+# ── Entry point ───────────────────────────────────────────────────────────────
+
+
+def main():
+    parser = argparse.ArgumentParser(description="OMX arm reset / grab script")
+    parser.add_argument("--port", default="/dev/ttyACM1")
+    parser.add_argument("--robot_id", default="omx_follower")
+    parser.add_argument("--mode", choices=["grab", "grab_and_place"], default="grab_and_place")
+    args = parser.parse_args()
+
+    logging.basicConfig(level=logging.INFO, format="%(levelname)s: %(message)s")
+
+    robot = OmxFollower(OmxFollowerConfig(port=args.port, id=args.robot_id))
+    robot.connect(calibrate=True)
+
+    try:
+        if args.mode == "grab":
+            grab_cube(robot)
+        elif args.mode == "grab_and_place":
+            grab_cube(robot)
+            place_cube(robot)
+
+    finally:
+        robot.disconnect()
+
+
+if __name__ == "__main__":
+    main()

examples/tutorial/rl/hilserl_example.py

@@ -4,13 +4,13 @@ from pathlib import Path
 from queue import Empty, Full
 
 import torch
-import torch.optim as optim
 
 from lerobot.datasets import LeRobotDataset
 from lerobot.envs.configs import HILSerlProcessorConfig, HILSerlRobotEnvConfig
-from lerobot.policies import SACConfig
-from lerobot.policies.sac.modeling_sac import SACPolicy
-from lerobot.policies.sac.reward_model.modeling_classifier import Classifier
+from lerobot.policies import GaussianActorConfig
+from lerobot.policies.gaussian_actor.modeling_gaussian_actor import GaussianActorPolicy
+from lerobot.rewards.classifier.modeling_classifier import Classifier
+from lerobot.rl.algorithms.sac import SACAlgorithm, SACAlgorithmConfig
 from lerobot.rl.buffer import ReplayBuffer
 from lerobot.rl.gym_manipulator import make_robot_env
 from lerobot.robots.so_follower import SO100FollowerConfig
@@ -28,7 +28,7 @@ def run_learner(
     transitions_queue: mp.Queue,
     parameters_queue: mp.Queue,
     shutdown_event: mp.Event,
-    policy_learner: SACPolicy,
+    policy_learner: GaussianActorPolicy,
     online_buffer: ReplayBuffer,
     offline_buffer: ReplayBuffer,
     lr: float = 3e-4,
@@ -40,8 +40,9 @@ def run_learner(
     policy_learner.train()
     policy_learner.to(device)
 
-    # Create Adam optimizer from scratch - simple and clean
-    optimizer = optim.Adam(policy_learner.parameters(), lr=lr)
+    algo_config = SACAlgorithmConfig.from_policy_config(policy_learner.config)
+    algorithm = SACAlgorithm(policy=policy_learner, config=algo_config)
+    algorithm.make_optimizers_and_scheduler()
 
     print(f"[LEARNER] Online buffer capacity: {online_buffer.capacity}")
     print(f"[LEARNER] Offline buffer capacity: {offline_buffer.capacity}")
@@ -83,24 +84,26 @@ def run_learner(
                 else:
                     batch[key] = online_batch[key]
 
-            loss, _ = policy_learner.forward(batch)
+            def batch_iter(b=batch):
+                while True:
+                    yield b
 
-            optimizer.zero_grad()
-            loss.backward()
-            optimizer.step()
+            stats = algorithm.update(batch_iter())
             training_step += 1
 
             if training_step % LOG_EVERY == 0:
+                log_dict = stats.to_log_dict()
                 print(
-                    f"[LEARNER] Training step {training_step}, Loss: {loss.item():.4f}, "
+                    f"[LEARNER] Training step {training_step}, "
+                    f"critic_loss: {log_dict.get('critic', 'N/A'):.4f}, "
                     f"Buffers: Online={len(online_buffer)}, Offline={len(offline_buffer)}"
                 )
 
             # Send updated parameters to actor every 10 training steps
             if training_step % SEND_EVERY == 0:
                 try:
-                    state_dict = {k: v.cpu() for k, v in policy_learner.state_dict().items()}
-                    parameters_queue.put_nowait(state_dict)
+                    weights = algorithm.get_weights()
+                    parameters_queue.put_nowait(weights)
                     print("[LEARNER] Sent updated parameters to actor")
                 except Full:
                     # Missing write due to queue not being consumed (should happen rarely)
@@ -113,7 +116,7 @@ def run_actor(
     transitions_queue: mp.Queue,
     parameters_queue: mp.Queue,
     shutdown_event: mp.Event,
-    policy_actor: SACPolicy,
+    policy_actor: GaussianActorPolicy,
     reward_classifier: Classifier,
     env_cfg: HILSerlRobotEnvConfig,
     device: torch.device = "mps",
@@ -144,15 +147,15 @@ def run_actor(
 
             while step < MAX_STEPS_PER_EPISODE and not shutdown_event.is_set():
                 try:
-                    new_params = parameters_queue.get_nowait()
-                    policy_actor.load_state_dict(new_params)
+                    new_weights = parameters_queue.get_nowait()
+                    policy_actor.load_state_dict(new_weights)
                     print("[ACTOR] Updated policy parameters from learner")
                 except Empty:  # No new updated parameters available from learner, waiting
                     pass
 
-                # Get action from policy
+                # Get action from policy (returns full action: continuous + discrete)
                 policy_obs = make_policy_obs(obs, device=device)
-                action_tensor = policy_actor.select_action(policy_obs)  # predicts a single action
+                action_tensor = policy_actor.select_action(policy_obs)
                 action = action_tensor.squeeze(0).cpu().numpy()
 
                 # Step environment
@@ -261,14 +264,14 @@ def main():
     action_features = hw_to_dataset_features(env.robot.action_features, "action")
 
     # Create SAC policy for action selection
-    policy_cfg = SACConfig(
+    policy_cfg = GaussianActorConfig(
         device=device,
         input_features=obs_features,
         output_features=action_features,
     )
 
-    policy_actor = SACPolicy(policy_cfg)
-    policy_learner = SACPolicy(policy_cfg)
+    policy_actor = GaussianActorPolicy(policy_cfg)
+    policy_learner = GaussianActorPolicy(policy_cfg)
 
     demonstrations_repo_id = "lerobot/example_hil_serl_dataset"
     offline_dataset = LeRobotDataset(repo_id=demonstrations_repo_id)

examples/tutorial/rl/reward_classifier_example.py

@@ -1,7 +1,7 @@
 import torch
 
 from lerobot.datasets import LeRobotDataset
-from lerobot.policies import RewardClassifierConfig, make_policy, make_pre_post_processors
+from lerobot.rewards import RewardClassifierConfig, make_reward_model, make_reward_pre_post_processors
 
 
 def main():
@@ -22,10 +22,10 @@ def main():
         model_name="microsoft/resnet-18",
     )
 
-    # Make policy, preprocessor, and optimizer
-    policy = make_policy(config, ds_meta=dataset.meta)
-    optimizer = config.get_optimizer_preset().build(policy.parameters())
-    preprocessor, _ = make_pre_post_processors(policy_cfg=config, dataset_stats=dataset.meta.stats)
+    # Make reward model, preprocessor, and optimizer
+    reward_model = make_reward_model(config, dataset_stats=dataset.meta.stats)
+    optimizer = config.get_optimizer_preset().build(reward_model.parameters())
+    preprocessor, _ = make_reward_pre_post_processors(config, dataset_stats=dataset.meta.stats)
 
     classifier_id = "<user>/reward_classifier_hil_serl_example"
 
@@ -42,7 +42,7 @@ def main():
             batch = preprocessor(batch)
 
             # Forward pass
-            loss, output_dict = policy.forward(batch)
+            loss, output_dict = reward_model.forward(batch)
 
             # Backward pass and optimization
             optimizer.zero_grad()
@@ -58,8 +58,8 @@ def main():
 
     print("Training finished!")
 
-    # You can now save the trained policy.
-    policy.push_to_hub(classifier_id)
+    # You can now save the trained reward model.
+    reward_model.push_to_hub(classifier_id)
 
 
 if __name__ == "__main__":

pyproject.toml

@@ -59,8 +59,8 @@ keywords = ["lerobot", "huggingface", "robotics",  "machine learning", "artifici
 
 dependencies = [
     # Core ML
-    "torch>=2.7,<2.11.0",
-    "torchvision>=0.22.0,<0.26.0",
+    "torch>=2.7,<2.12.0",
+    "torchvision>=0.22.0,<0.27.0",
     "numpy>=2.0.0,<2.3.0", # NOTE: Explicitly listing numpy helps the resolver converge faster. Upper bound imposed by opencv-python-headless.
     "opencv-python-headless>=4.9.0,<4.14.0",
     "Pillow>=10.0.0,<13.0.0",
@@ -95,11 +95,22 @@ dependencies = [
 
 # ── Feature-scoped extras ──────────────────────────────────
 dataset = [
-    "datasets>=4.0.0,<5.0.0",
+    "datasets>=4.7.0,<5.0.0",
     "pandas>=2.0.0,<3.0.0", # NOTE: Transitive dependency of datasets
     "pyarrow>=21.0.0,<30.0.0", # NOTE: Transitive dependency of datasets
     "lerobot[av-dep]",
-    "torchcodec>=0.3.0,<0.11.0; sys_platform != 'win32' and (sys_platform != 'linux' or (platform_machine != 'aarch64' and platform_machine != 'arm64' and platform_machine != 'armv7l')) and (sys_platform != 'darwin' or platform_machine != 'x86_64')", # NOTE: Windows support starts at version 0.7 (needs torch==2.8), ffmpeg>=8 support starts at version 0.8.1 (needs torch==2.9), system-wide ffmpeg support starts at version 0.10 (needs torch==2.10).
+
+    # NOTE: torchcodec wheel availability matrix (PyPI):
+    #   - linux x86_64/amd64 + macOS arm64 : wheels since 0.3.0 (the historic supported set).
+    #   - win32 x86_64                     : wheels since 0.7.0  (needs torch>=2.8).
+    #   - linux aarch64/arm64              : wheels since 0.11.0 (needs torch>=2.11).
+    #   - macOS x86_64 (Intel) and linux armv7l: no wheels in any released version -> fall through to the PyAV decoder.
+    # Each platform gets its own line so the resolver picks the minimum version that has a wheel for it.
+
+    # Other torch/torchcodec pairings (informational): 0.8.1 = ffmpeg>=8 support, 0.10 = system-wide ffmpeg support, 0.12 needs torch==2.12.
+    "torchcodec>=0.3.0,<0.12.0; (sys_platform == 'linux' and (platform_machine == 'x86_64' or platform_machine == 'AMD64')) or (sys_platform == 'darwin' and platform_machine == 'arm64')",
+    "torchcodec>=0.7.0,<0.12.0; sys_platform == 'win32'",
+    "torchcodec>=0.11.0,<0.12.0; sys_platform == 'linux' and (platform_machine == 'aarch64' or platform_machine == 'arm64')",
     "jsonlines>=4.0.0,<5.0.0",
 ]
 training = [
@@ -127,8 +138,10 @@ dataset_viz = ["lerobot[dataset]", "lerobot[viz]"]
 # Common
 av-dep = ["av>=15.0.0,<16.0.0"]
 pygame-dep = ["pygame>=2.5.1,<2.7.0"]
-placo-dep = ["placo>=0.9.6,<0.9.17"]
-transformers-dep = ["transformers==5.3.0"] # TODO(Steven): https://github.com/huggingface/lerobot/pull/3249
+# NOTE: 0.9.16 links against liburdfdom_sensor.so.4, which is unavailable on Ubuntu 24.04
+# (noble ships urdfdom 3.x). Cap below 0.9.16 until system urdfdom 4.x is broadly available.
+placo-dep = ["placo>=0.9.6,<0.9.16"]
+transformers-dep = ["transformers>=5.4.0,<5.6.0"]
 grpcio-dep = ["grpcio==1.73.1", "protobuf>=6.31.1,<6.32.0"]
 can-dep = ["python-can>=4.2.0,<5.0.0"]
 peft-dep = ["peft>=0.18.0,<1.0.0"]
@@ -140,6 +153,8 @@ pyserial-dep = ["pyserial>=3.5,<4.0"]
 deepdiff-dep = ["deepdiff>=7.0.1,<9.0.0"]
 pynput-dep = ["pynput>=1.7.8,<1.9.0"]
 pyzmq-dep = ["pyzmq>=26.2.1,<28.0.0"]
+motorbridge-dep = ["motorbridge>=0.3.2,<0.4.0"]
+motorbridge-smart-servo-dep = ["motorbridge-smart-servo>=0.0.4,<0.1.0"]
 
 # Motors
 feetech = ["feetech-servo-sdk>=1.0.0,<2.0.0", "lerobot[pyserial-dep]", "lerobot[deepdiff-dep]"]
@@ -163,6 +178,9 @@ unitree_g1 = [
     "lerobot[pygame-dep]",
 ]
 reachy2 = ["reachy2_sdk>=1.0.15,<1.1.0"]
+# Seeed Studio reBot B601-DM follower (motorbridge / CAN) + StarArm102 / reBot Arm 102
+# leader (motorbridge-smart-servo / FashionStar UART servos).
+rebot = ["lerobot[motorbridge-dep]", "lerobot[motorbridge-smart-servo-dep]"]
 kinematics = ["lerobot[placo-dep]"]
 intelrealsense = [
     "pyrealsense2>=2.55.1.6486,<2.57.0 ; sys_platform != 'darwin'",
@@ -180,6 +198,7 @@ wallx = [
     "lerobot[qwen-vl-utils-dep]",
 ]
 pi = ["lerobot[transformers-dep]", "lerobot[scipy-dep]"]
+molmoact2 = ["lerobot[transformers-dep]", "lerobot[peft-dep]", "lerobot[scipy-dep]"]
 smolvla = ["lerobot[transformers-dep]", "num2words>=0.5.14,<0.6.0", "accelerate>=1.7.0,<2.0.0"]
 multi_task_dit = ["lerobot[transformers-dep]", "lerobot[diffusers-dep]"]
 groot = [
@@ -193,13 +212,28 @@ groot = [
     "flash-attn>=2.5.9,<3.0.0 ; sys_platform != 'darwin'"
 ]
 sarm = ["lerobot[transformers-dep]", "pydantic>=2.0.0,<3.0.0", "faker>=33.0.0,<35.0.0", "lerobot[matplotlib-dep]", "lerobot[qwen-vl-utils-dep]"]
+robometer = ["lerobot[transformers-dep]", "lerobot[qwen-vl-utils-dep]", "lerobot[peft-dep]"]
+topreward = ["lerobot[transformers-dep]"]
 xvla = ["lerobot[transformers-dep]"]
-hilserl = ["lerobot[transformers-dep]", "gym-hil>=0.1.13,<0.2.0", "lerobot[grpcio-dep]", "lerobot[placo-dep]"]
+eo1 = ["lerobot[transformers-dep]", "lerobot[qwen-vl-utils-dep]"]
+hilserl = ["lerobot[transformers-dep]", "lerobot[dataset]", "gym-hil>=0.1.13,<0.2.0", "lerobot[grpcio-dep]", "lerobot[placo-dep]"]
 
 # Features
 async = ["lerobot[grpcio-dep]", "lerobot[matplotlib-dep]"]
 peft = ["lerobot[transformers-dep]", "lerobot[peft-dep]"]
 
+# Annotation pipeline (lerobot-annotate). vllm is the preferred backend
+# on Linux, with a transformers fallback elsewhere; openai is the default
+# backend and talks to any OpenAI-compatible server (``vllm serve`` /
+# ``transformers serve`` / hosted endpoints). Distributed execution is
+# delegated to Hugging Face Jobs (see examples/annotations/run_hf_job.py).
+annotations = [
+    "lerobot[dataset]",
+    "lerobot[transformers-dep]",
+    "openai>=1.40,<2.0",
+    "vllm>=0.6.0,<1.0.0; sys_platform == 'linux'",
+]
+
 # Development
 dev = ["pre-commit>=3.7.0,<5.0.0", "debugpy>=1.8.1,<1.9.0", "lerobot[grpcio-dep]", "grpcio-tools==1.73.1", "mypy>=1.19.1", "ruff>=0.14.1", "lerobot[notebook]"]
 notebook = ["jupyter>=1.0.0,<2.0.0", "ipykernel>=6.0.0,<7.0.0"]
@@ -248,12 +282,14 @@ all = [
     "lerobot[lekiwi]",
     "lerobot[openarms]",
     "lerobot[reachy2]",
+    "lerobot[rebot]",
     "lerobot[kinematics]",
     "lerobot[intelrealsense]",
     "lerobot[diffusion]",
     "lerobot[multi_task_dit]",
     "lerobot[wallx]",
     "lerobot[pi]",
+    "lerobot[molmoact2]",
     "lerobot[smolvla]",
     # "lerobot[groot]", TODO(Steven): Gr00t requires specific installation instructions for flash-attn
     "lerobot[xvla]",
@@ -268,6 +304,8 @@ all = [
     "lerobot[libero]; sys_platform == 'linux'",
     "lerobot[metaworld]",
     "lerobot[sarm]",
+    "lerobot[robometer]",
+    "lerobot[topreward]",
     "lerobot[peft]",
     # "lerobot[unitree_g1]", TODO: Unitree requires specific installation instructions for unitree_sdk2
 ]
@@ -289,11 +327,26 @@ lerobot-find-joint-limits="lerobot.scripts.lerobot_find_joint_limits:main"
 lerobot-imgtransform-viz="lerobot.scripts.lerobot_imgtransform_viz:main"
 lerobot-edit-dataset="lerobot.scripts.lerobot_edit_dataset:main"
 lerobot-setup-can="lerobot.scripts.lerobot_setup_can:main"
+lerobot-annotate="lerobot.scripts.lerobot_annotate:main"
 lerobot-rollout="lerobot.scripts.lerobot_rollout:main"
 
 # ---------------- Tool Configurations ----------------
+
+# cu128 wheels keep broad hardware reach; the driver floor is 570.86.
+# To use a different CUDA variant, reinstall torch with an explicit index, e.g.:
+#   uv pip install --force-reinstall torch torchvision \
+#       --index-url https://download.pytorch.org/whl/cu130
+[[tool.uv.index]]
+name = "pytorch-cu128"
+url = "https://download.pytorch.org/whl/cu128"
+explicit = true
+
+[tool.uv.sources]
+torch = [{ index = "pytorch-cu128", marker = "sys_platform == 'linux'" }]
+torchvision = [{ index = "pytorch-cu128", marker = "sys_platform == 'linux'" }]
+
 [tool.setuptools.package-data]
-lerobot = ["envs/*.json"]
+lerobot = ["envs/*.json", "annotations/steerable_pipeline/prompts/*.txt"]
 
 [tool.setuptools.packages.find]
 where = ["src"]
@@ -354,7 +407,7 @@ exclude_dirs = [
     "benchmarks",
     "src/lerobot/datasets/push_dataset_to_hub",
 ]
-skips = ["B101", "B311", "B404", "B603", "B615"]
+skips = ["B101", "B311", "B404", "B603", "B607", "B615"]
 
 [tool.typos]
 default.extend-ignore-re = [
@@ -369,8 +422,11 @@ default.extend-ignore-identifiers-re = [
     "ein",
     "thw",
     "inpt",
+    "arange",
+    "is_compileable",
     "ROBOTIS",
-    "OT_VALUE"
+    "OT_VALUE",
+    "VanderBilt"
 ]
 
 # TODO: Uncomment when ready to use

src/lerobot/annotations/__init__.py

@@ -0,0 +1,15 @@
+#!/usr/bin/env python
+
+# Copyright 2026 The HuggingFace Inc. team. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.

src/lerobot/annotations/steerable_pipeline/__init__.py

@@ -0,0 +1,36 @@
+#!/usr/bin/env python
+
+# Copyright 2026 The HuggingFace Inc. team. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""Steerable annotation pipeline producing ``language_persistent`` and
+``language_events`` columns for LeRobot datasets.
+
+The pipeline is decomposed into three independently runnable modules whose
+outputs are staged per-episode before a final parquet rewrite:
+
+- :mod:`.modules.plan_subtasks_memory` (the ``plan`` module) — persistent styles
+- :mod:`.modules.interjections_and_speech` (the ``interjections`` module) — event styles + speech
+- :mod:`.modules.general_vqa` (the ``vqa`` module) — event-style VQA pairs
+"""
+
+from .config import AnnotationPipelineConfig
+from .validator import StagingValidator, ValidationReport
+from .writer import LanguageColumnsWriter
+
+__all__ = [
+    "AnnotationPipelineConfig",
+    "LanguageColumnsWriter",
+    "StagingValidator",
+    "ValidationReport",
+]

src/lerobot/annotations/steerable_pipeline/config.py

@@ -0,0 +1,412 @@
+#!/usr/bin/env python
+
+# Copyright 2026 The HuggingFace Inc. team. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from __future__ import annotations
+
+from dataclasses import dataclass, field
+from pathlib import Path
+from typing import Any
+
+
+@dataclass
+class PlanConfig:
+    """``plan`` module: plan + subtasks + memory + task augmentation.
+
+    The ``plan`` module attaches the whole episode as one Qwen-VL video
+    block; ``max_video_frames`` only caps the frames packed in (a
+    model-capacity bound, not an annotation-logic knob).
+    """
+
+    enabled: bool = True
+
+    # Number of ``task_aug`` rephrasings emitted at ``t=0``. The renderer's
+    # ``${task}`` binding rotates among them per ``sample_idx``. ``0`` disables.
+    n_task_rephrasings: int = 10
+
+    # When to derive the task from the video instead of using
+    # ``record.episode_task``: ``off``, ``if_short`` (short / placeholder /
+    # missing canonical task), or ``always``. The derived task replaces the
+    # canonical one for every ``plan``-module prompt; ``meta/tasks.parquet``
+    # is never modified.
+    derive_task_from_video: str = "if_short"
+    derive_task_min_words: int = 3
+
+    # Frame sampling for the subtask-decomposition prompt. Frames are
+    # sampled uniformly across the whole episode up to ``max_video_frames``
+    # (so longer episodes are subsampled, not truncated).
+    #
+    # ``max_video_frames`` is a HARD context-budget cap. With the embedded-
+    # frame path (use_video_url=false), every frame becomes ~250-320 vision
+    # tokens, so 128 frames ≈ 33-39k tokens — over a 32k-context VLM. 32
+    # frames (~8-10k tokens) leaves ample room for the prompt + the
+    # describe / verify passes. Raise only if your serving context is
+    # larger AND your episodes need finer temporal resolution; if you hit
+    # "Input length exceeds maximum context length", lower this.
+    frames_per_second: float = 1.0
+    max_video_frames: int = 32
+
+    # Windowed subtask generation for CONSTANT temporal density. When > 0
+    # and an episode is longer than this many seconds, the plan module
+    # processes the episode in consecutive windows of this length, each
+    # sampled at ``frames_per_second``, instead of subsampling the whole
+    # episode to ``max_video_frames`` (which makes long episodes sparse).
+    # The describe -> segment -> verify chain runs per window; results are
+    # offset to absolute time, merged, and stitched into a contiguous
+    # whole-episode cover. Cost scales with episode length (≈ chain calls
+    # × ceil(duration / window)). Set to ~max_video_frames / frames_per_
+    # second (e.g. 32s at 1 fps) so each window fills — but never exceeds —
+    # the per-call frame budget. 0 disables (single whole-episode call).
+    subtask_window_seconds: float = 0.0
+
+    min_subtask_seconds: float = 1.5
+    plan_max_steps: int = 8
+
+    # ``subtask_describe_first``: run a grounding pass that narrates ONLY
+    # what is visible in the video (no subtask JSON yet), then inject that
+    # description into the segmentation prompt. Forces the model to observe
+    # before committing to structured output — the strongest lever against
+    # subtasks invented from the task text. ON by default; +1 VLM call/ep.
+    # Set False to trade quality for fewer calls on easy datasets.
+    subtask_describe_first: bool = True
+
+    # Emit ``style="plan"`` rows (the numbered still-todo list re-emitted at
+    # every subtask boundary). Set False to keep only subtasks + memory and
+    # skip the plan rows entirely — saves one ``_generate_plan`` VLM call per
+    # subtask boundary. Subtask and memory generation are unaffected.
+    emit_plan: bool = True
+
+    # NOTE: subtask spans are ALWAYS stitched into a contiguous
+    # full-episode cover (first subtask pulled back to t0, gaps closed,
+    # last span extended to t_last) as a deterministic post-step in
+    # ``_generate_subtasks._stitch_full_coverage``. This is not
+    # configurable — a sparse / gap-ridden subtask timeline is never
+    # desirable for conditioning, so it is unconditional.
+
+    # When True (and backend supports it, e.g. ``openai``), the ``plan``
+    # module sends a ``video_url`` block pointing at a per-episode mp4
+    # subclip and lets the server sample frames at ``use_video_url_fps``.
+    use_video_url: bool = False
+    use_video_url_fps: float = 1.0
+
+    # Structured per-subtask action records (Phase 1a + 1b, inspired by
+    # EgoMimic's annotator form). For each generated subtask span, the
+    # VLM extracts a typed record (verb / object / arm / grasp_type /
+    # destination / mistake). A deterministic Python template renders
+    # that record back to canonical subtask text — reducing the VLM's
+    # "creative" surface to just the perception step. See
+    # ``ActionRecordsConfig`` for details. Off by default (back-compat).
+    action_records: ActionRecordsConfig = field(default_factory=lambda: ActionRecordsConfig())
+
+    # Structured 5-axis augmentation taxonomy for the t=0 task variants
+    # (replaces the free-form ``n_task_rephrasings`` flow when enabled).
+    # Mirrors EgoMimic's ``augment_prompt.txt`` taxonomy: instead of N
+    # free-form rephrasings, the VLM produces variants along named
+    # axes (synonym / omit_arm / omit_orientation / omit_grasp_method /
+    # combined). Off by default (back-compat).
+    task_aug_axes: TaskAugAxesConfig = field(default_factory=lambda: TaskAugAxesConfig())
+
+
+@dataclass
+class ActionRecordsConfig:
+    """Structured per-subtask action record extraction.
+
+    When ``enabled=True``, after the existing subtask-span generation in
+    ``plan_subtasks_memory.py``, the module makes one extra VLM call per
+    subtask to extract a typed record::
+
+        {
+            "verb": "pick" | "place" | "press" | ...,  # closed vocabulary
+            "object": "<canonical_object_name>",
+            "arm": "left" | "right" | "both" | null,
+            "grasp_type": "pinch" | "wrap" | "hook" | ... | null,
+            "destination": "<canonical_destination>" | null,
+            "mistake": "<short text>" | null,
+        }
+
+    The record is emitted as a separate row with ``style="action_record"``
+    (``content=json.dumps(record)``) at the subtask's start timestamp.
+    It is PURELY ADDITIVE — it never touches the VLM's subtask text.
+    Downstream training can consume the typed schema directly (e.g.
+    auxiliary supervision on verb / arm / grasp classification heads)
+    while the subtask string the policy conditions on stays exactly what
+    the subtask module produced. (Reconstructing subtask text from these
+    fields was too easy for the VLM to hallucinate on tasks that don't
+    fit the manipulation schema — navigation tasks yielded nonsense like
+    ``move stove to stove`` — so that path was removed.)
+
+    Cost: one extra VLM call per subtask. For an 8-subtask episode this
+    means ~8x more VLM calls in the plan module — still cheap relative
+    to the action-expert training cost, but worth knowing.
+    """
+
+    enabled: bool = False
+
+    # When True (default), emit a separate row with ``style="action_record"``
+    # and ``content=json.dumps(record)`` at the subtask's start timestamp.
+    # This is the only output of the feature — set ``enabled=False`` to
+    # skip the extra VLM calls entirely.
+    emit_record_row: bool = True
+
+    # Frame sampling for the per-subtask VLM call (similar to the
+    # interjection module's window). Anchored to the subtask span.
+    frames_per_subtask: int = 4
+
+    # Closed verb vocabulary. The prompt instructs the VLM to pick
+    # exactly one. Override per-dataset (e.g. ``["pick", "place", "open",
+    # "close"]`` for door-only manipulation) for tighter constraint.
+    verb_vocabulary: tuple[str, ...] = (
+        "pick",
+        "place",
+        "push",
+        "pull",
+        "open",
+        "close",
+        "turn",
+        "press",
+        "lift",
+        "insert",
+        "pour",
+        "move",
+        "reach",
+        "grasp",
+        "release",
+        "wipe",
+        "dump",
+    )
+
+    # Closed grasp-type vocabulary. ``null`` is always allowed (no
+    # contact / unclear). Adjust per-hardware (e.g. drop ``hook`` /
+    # ``key`` for parallel-jaw grippers).
+    grasp_vocabulary: tuple[str, ...] = (
+        "pinch",
+        "wrap",
+        "hook",
+        "key",
+        "lateral",
+    )
+
+
+@dataclass
+class TaskAugAxesConfig:
+    """Structured 5-axis augmentation taxonomy for t=0 task variants.
+
+    When ``enabled=True``, replaces the free-form ``n_task_rephrasings``
+    flow with a structured prompt that produces variants along five
+    named axes (mirroring EgoMimic's ``augment_prompt.txt``):
+
+      * ``synonym_paraphrase`` — different wording / verbs, all
+        information preserved.
+      * ``omit_arm`` — drop the left/right/both arm specification.
+      * ``omit_orientation`` — drop orientation cues (upright,
+        sideways, ...).
+      * ``omit_grasp_method`` — drop grip / grasp method specification.
+      * ``combined_omissions`` — combine two of the above
+        simultaneously.
+
+    Default counts (3+3+2+2+2 = 12 variants per task) match EgoMimic.
+    Axes that have nothing to omit in the source task (e.g. ``omit_arm``
+    when the task doesn't mention an arm) emit fewer entries rather
+    than pad — the prompt instructs the VLM accordingly.
+
+    Each variant is emitted as a ``task_aug`` row at ``t=0`` (same
+    style as the free-form variants), so the rest of the pipeline /
+    training recipe doesn't need to know about the taxonomy.
+    """
+
+    enabled: bool = False
+
+    synonym_paraphrase: int = 3
+    omit_arm: int = 3
+    omit_orientation: int = 2
+    omit_grasp_method: int = 2
+    combined_omissions: int = 2
+
+    @property
+    def total(self) -> int:
+        """Sum of requested variants across all axes (upper bound)."""
+        return (
+            self.synonym_paraphrase
+            + self.omit_arm
+            + self.omit_orientation
+            + self.omit_grasp_method
+            + self.combined_omissions
+        )
+
+
+@dataclass
+class InterjectionsConfig:
+    """``interjections`` module: interjections + paired speech."""
+
+    enabled: bool = True
+
+    # Each interjection emits a paired ``(interjection, speech)`` event row
+    # and triggers a ``plan`` refresh at the same timestamp via the
+    # ``plan`` module.
+    max_interjections_per_episode: int = 3
+    interjection_min_t: float = 2.0
+
+    # Visual context attached to the interjection prompt: a short window
+    # of frames centered on the chosen timestamp so the VLM sees the
+    # ongoing motion rather than a single frozen frame.
+    interjection_window_seconds: float = 2.0
+    interjection_window_frames: int = 4
+
+
+@dataclass
+class VqaConfig:
+    """``vqa`` module: general VQA."""
+
+    enabled: bool = True
+    vqa_emission_hz: float = 1.0
+    K: int = 1
+    """How many *consecutive* frames each emission tick anchors a VQA pair
+    to. The VLM grounds its answer (bbox / keypoint coordinates, count, …)
+    against the *first* anchored frame's image, so anchoring K>1 frames
+    copies that same answer onto later frames where the scene has already
+    moved — stale labels. Default ``1``: a VQA pair lands on exactly its
+    emission frame, no temporal smear. Raise it only to trade label
+    precision for more (noisier) VQA frames."""
+    question_types: tuple[str, ...] = ("bbox", "keypoint", "count", "attribute", "spatial")
+
+    # Camera restriction. By default VQA iterates EVERY camera the
+    # dataset declares (one VQA pair per camera per emission tick). Set
+    # ``restrict_to_default_camera=True`` to ground VQA on only the
+    # single ``--vlm.camera_key`` stream — the same camera the plan /
+    # interjection modules use — so the whole pipeline focuses on one
+    # view. Use this when you want every annotation grounded on, e.g.,
+    # ``observation.images.base`` and nothing else.
+    restrict_to_default_camera: bool = False
+
+
+@dataclass
+class VlmConfig:
+    """Shared Qwen-VL client configuration."""
+
+    # One of ``vllm``, ``transformers``, ``openai``, or ``stub`` (tests).
+    # ``openai`` talks to a local OpenAI-compatible server; the CLI
+    # auto-spawns one when ``auto_serve=True``.
+    backend: str = "openai"
+    model_id: str = "Qwen/Qwen3.6-35B-A3B-FP8"
+
+    # OpenAI-compatible server endpoint; ``EMPTY`` works for local servers.
+    api_base: str = "http://localhost:8000/v1"
+    api_key: str = "EMPTY"
+
+    # When True with ``backend=openai``, the CLI probes ``api_base`` and
+    # spawns a server if none answers (default: ``transformers serve``).
+    # Set to False to fail fast when pointing at a remote endpoint.
+    auto_serve: bool = True
+    serve_port: int = 8000
+    # Override the auto-serve command. ``{port}`` is substituted per replica
+    # when ``parallel_servers > 1``.
+    serve_command: str | None = None
+
+    # Run multiple independent inference servers for round-robin client
+    # routing (each pinned to a GPU via ``CUDA_VISIBLE_DEVICES`` and bound
+    # to ``serve_port + i``). ``num_gpus=0`` means one GPU per replica.
+    parallel_servers: int = 1
+    num_gpus: int = 0
+    client_concurrency: int = 16
+    serve_ready_timeout_s: float = 600.0
+
+    max_new_tokens: int = 512
+    temperature: float = 0.2
+    json_mode: bool = True
+    batch_size: int = 4
+    tensor_parallel_size: int = 1
+
+    # Fraction of GPU memory vllm allocates for weights + KV cache.
+    gpu_memory_utilization: float = 0.9
+    # Cap context length (None = model default). On 80 GB H100 a 30B BF16
+    # model often needs <= 8192 to leave KV-cache headroom.
+    max_model_len: int | None = None
+    trust_remote_code: bool = False
+
+    # Override the camera stream used for keyframe attachment. None picks
+    # the first ``observation.images.*`` key the dataset declares.
+    camera_key: str | None = None
+    # Forwarded as ``extra_body.chat_template_kwargs`` on every chat call;
+    # use to pass model-specific flags such as ``{"enable_thinking": false}``.
+    chat_template_kwargs: dict[str, Any] | None = None
+
+
+@dataclass
+class ExecutorConfig:
+    """Executor settings.
+
+    Distributed execution is provided by Hugging Face Jobs (see
+    ``examples/annotation/run_hf_job.py``); this config only controls
+    intra-process episode concurrency.
+    """
+
+    # Episodes processed concurrently within each module phase. Each
+    # in-flight episode dispatches 3-5 dependent VLM calls, so this is the
+    # main knob for saturating ``parallel_servers`` and ``client_concurrency``.
+    episode_parallelism: int = 16
+
+
+@dataclass
+class AnnotationPipelineConfig:
+    """Top-level config for ``lerobot-annotate``.
+
+    The writer rewrites ``data/chunk-*/file-*.parquet`` in place. Multiple
+    revisions of the same dataset live in separate copies.
+    """
+
+    # Hub dataset id. Used as the download source when ``root`` is unset,
+    # and as the destination repo when ``push_to_hub`` is enabled and
+    # ``dest_repo_id`` is unset.
+    repo_id: str | None = None
+
+    # Optional separate Hub dataset id to push the annotated result to. When
+    # unset, ``push_to_hub`` uploads back to ``repo_id`` (annotate in place);
+    # when set, the source ``repo_id`` is left untouched.
+    dest_repo_id: str | None = None
+
+    root: Path | None = None
+
+    # Defaults to ``<root>/.annotate_staging/`` when unset.
+    staging_dir: Path | None = None
+
+    seed: int = 1729
+
+    plan: PlanConfig = field(default_factory=PlanConfig)
+    interjections: InterjectionsConfig = field(default_factory=InterjectionsConfig)
+    vqa: VqaConfig = field(default_factory=VqaConfig)
+
+    vlm: VlmConfig = field(default_factory=VlmConfig)
+    executor: ExecutorConfig = field(default_factory=ExecutorConfig)
+
+    skip_validation: bool = False
+    only_episodes: tuple[int, ...] | None = None
+
+    # Keyframe decode backend. When unset, the pipeline decodes with the
+    # ffmpeg CLI: it decodes AV1 and runs each decode as an isolated child
+    # process, which is both crash-safe and safe under the concurrent
+    # decode the executor performs (torchcodec is not thread-safe and
+    # SIGSEGVs there). Set to ``"torchcodec"`` or ``"pyav"`` to pin an
+    # in-process decoder when its build is known thread-safe.
+    video_backend: str | None = None
+
+    # When True, upload the annotated dataset to the Hugging Face Hub:
+    # to ``dest_repo_id`` if set, otherwise back to ``repo_id``. One of
+    # the two must be set for this to take effect.
+    push_to_hub: bool = False
+    push_private: bool = False
+    push_commit_message: str | None = None
+
+    def resolved_staging_dir(self, root: Path) -> Path:
+        return self.staging_dir if self.staging_dir is not None else root / ".annotate_staging"

src/lerobot/annotations/steerable_pipeline/executor.py

@@ -0,0 +1,259 @@
+#!/usr/bin/env python
+
+# Copyright 2026 The HuggingFace Inc. team. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""In-process executor that runs the annotation phases.
+
+The executor plans **seven phases** in the dependency order from the plan:
+
+    phase 0: vocabulary discovery — derive a small canonical vocabulary
+             from the first few sample-episode videos (subtask labels +
+             memory milestones) and persist it next to the dataset; the
+             ``plan`` module then constrains every per-episode generation
+             to those strings, so the downstream policy sees a small,
+             repeatable conditioning distribution
+    phase 1: ``plan`` module (plan + subtasks + memory)
+    phase 2: ``interjections`` module (interjections + speech)
+    phase 3: ``plan`` plan-update pass — re-runs plan emission at every
+             interjection timestamp produced by phase 2
+    phase 4: ``vqa`` module (VQA)
+    phase 5: validator
+    phase 6: writer
+
+Phase 3 is why the ``plan`` module must be re-entered after the
+``interjections`` module — to refresh ``plan`` rows at interjection
+timestamps.
+
+Distributed execution is provided by Hugging Face Jobs (see
+``examples/annotations/run_hf_job.py``); the runner inside the job
+invokes ``lerobot-annotate`` which uses this in-process executor.
+Episode-level concurrency is controlled by
+``ExecutorConfig.episode_parallelism``.
+"""
+
+from __future__ import annotations
+
+import logging
+import time
+from concurrent.futures import ThreadPoolExecutor, as_completed
+from dataclasses import dataclass
+from pathlib import Path
+from typing import Any
+
+from .config import AnnotationPipelineConfig
+from .reader import EpisodeRecord, iter_episodes
+from .staging import EpisodeStaging
+from .validator import StagingValidator
+from .writer import LanguageColumnsWriter
+
+logger = logging.getLogger(__name__)
+
+
+@dataclass
+class PhaseResult:
+    """Summary of one pipeline phase across all episodes."""
+
+    name: str
+    episodes_processed: int
+    episodes_skipped: int
+
+
+@dataclass
+class PipelineRunSummary:
+    """Aggregated result returned by :meth:`Executor.run`."""
+
+    phases: list[PhaseResult]
+    written_paths: list[Path]
+    validation_report: Any  # ValidationReport, kept Any to avoid import cycle
+
+
+@dataclass
+class Executor:
+    """Run all six phases over a dataset root in-process.
+
+    Episode-level concurrency comes from ``ExecutorConfig.episode_parallelism``
+    (a thread pool); cluster-level concurrency comes from running this
+    executor inside a Hugging Face Job. Tests construct the executor
+    directly with stub modules.
+    """
+
+    config: AnnotationPipelineConfig
+    plan: Any  # PlanSubtasksMemoryModule
+    interjections: Any  # InterjectionsAndSpeechModule
+    vqa: Any  # GeneralVqaModule
+    writer: LanguageColumnsWriter
+    validator: StagingValidator
+
+    def run(self, root: Path) -> PipelineRunSummary:
+        records = list(iter_episodes(root, only_episodes=self.config.only_episodes))
+        n = len(records)
+        if n == 0:
+            raise ValueError(f"No episodes found under {root}/data/")
+
+        print(f"[annotate] {n} episodes total", flush=True)
+
+        staging_dir = self.config.resolved_staging_dir(root)
+        staging_dir.mkdir(parents=True, exist_ok=True)
+
+        phases: list[PhaseResult] = []
+
+        # Phase 1: ``plan`` module (plan + subtasks + memory)
+        phases.append(self._run_module_phase("plan", records, staging_dir, self.plan))
+        # Phase 2: ``interjections`` module (interjections + speech). It
+        # reads the ``plan`` module's subtask rows from the same staging
+        # tree to ground the interjection prompt in the correct local subtask.
+        phases.append(self._run_module_phase("interjections", records, staging_dir, self.interjections))
+        # Phase 3: ``plan`` plan-update pass at interjection timestamps.
+        phases.append(self._run_plan_update_phase(records, staging_dir))
+        # Phase 4: ``vqa`` module (VQA)
+        phases.append(self._run_module_phase("vqa", records, staging_dir, self.vqa))
+
+        print("[annotate] running validator...", flush=True)
+        report = self.validator.validate(records, staging_dir)
+        if not report.ok and not self.config.skip_validation:
+            raise RuntimeError(f"Staging validation failed: {report.summary()}")
+        print(f"[annotate] validator: {report.summary()}", flush=True)
+
+        print(f"[annotate] writing parquet shards into {root}/data/...", flush=True)
+        written = self.writer.write_all(records, staging_dir, root)
+        print(f"[annotate] wrote {len(written)} shard(s); pipeline complete", flush=True)
+
+        # Keep meta/info.json aligned with the parquet schema we just wrote.
+        # Idempotent and additive: existing user metadata is preserved.
+        self._ensure_annotation_metadata_in_info(root)
+
+        return PipelineRunSummary(phases=phases, written_paths=written, validation_report=report)
+
+    @staticmethod
+    def _ensure_annotation_metadata_in_info(root: Path) -> None:
+        """Write language features and canonical tools to ``meta/info.json``.
+
+        ``LanguageColumnsWriter`` adds ``language_persistent`` and
+        ``language_events`` to parquet shards. The metadata must advertise
+        those columns too, otherwise non-streaming ``LeRobotDataset`` loads
+        cast against the old schema and fail on the extra parquet columns.
+        """
+        from lerobot.datasets.io_utils import load_info, write_info  # noqa: PLC0415
+        from lerobot.datasets.language import SAY_TOOL_SCHEMA, language_feature_info  # noqa: PLC0415
+
+        info_path = root / "meta" / "info.json"
+        if not info_path.exists():
+            return
+        try:
+            info = load_info(root)
+        except Exception as exc:  # noqa: BLE001
+            print(f"[annotate] could not read {info_path}: {exc}", flush=True)
+            return
+
+        changed = False
+
+        merged_features = {**info.features, **language_feature_info()}
+        if merged_features != info.features:
+            info.features = merged_features
+            changed = True
+
+        existing = info.tools or []
+        names = {(t.get("function") or {}).get("name") for t in existing if isinstance(t, dict)}
+        if SAY_TOOL_SCHEMA["function"]["name"] not in names:
+            info.tools = [*existing, SAY_TOOL_SCHEMA]
+            changed = True
+
+        if changed:
+            write_info(info, root)
+            print(
+                "[annotate] meta/info.json: "
+                f"language_features={list(language_feature_info())}, "
+                f"tools={[t['function']['name'] for t in (info.tools or [])]}",
+                flush=True,
+            )
+
+    def _run_module_phase(
+        self,
+        name: str,
+        records: list[EpisodeRecord],
+        staging_dir: Path,
+        module: Any,
+    ) -> PhaseResult:
+        if not module.enabled:
+            print(f"[annotate] phase={name} skipped (module disabled)", flush=True)
+            return PhaseResult(name=name, episodes_processed=0, episodes_skipped=len(records))
+        n = len(records)
+        parallelism = max(1, min(self.config.executor.episode_parallelism, n))
+        print(
+            f"[annotate] phase={name} starting on {n} episode(s) (parallelism={parallelism})",
+            flush=True,
+        )
+        t0 = time.time()
+
+        def _do(idx_record: tuple[int, EpisodeRecord]) -> tuple[int, int, float]:
+            i, record = idx_record
+            ep_start = time.time()
+            staging = EpisodeStaging(staging_dir, record.episode_index)
+            module.run_episode(record, staging)
+            return i, record.episode_index, time.time() - ep_start
+
+        processed = 0
+        if parallelism == 1:
+            for i, record in enumerate(records, 1):
+                _, ep_idx, elapsed = _do((i, record))
+                processed += 1
+                print(
+                    f"[annotate]   {name} episode {i}/{n} (idx={ep_idx}) done in {elapsed:.1f}s",
+                    flush=True,
+                )
+        else:
+            with ThreadPoolExecutor(max_workers=parallelism) as pool:
+                futures = [pool.submit(_do, (i, r)) for i, r in enumerate(records, 1)]
+                for fut in as_completed(futures):
+                    i, ep_idx, elapsed = fut.result()
+                    processed += 1
+                    print(
+                        f"[annotate]   {name} episode {processed}/{n} "
+                        f"(idx={ep_idx}, submit_order={i}) done in {elapsed:.1f}s",
+                        flush=True,
+                    )
+        total = time.time() - t0
+        print(f"[annotate] phase={name} complete: {processed}/{n} in {total:.1f}s", flush=True)
+        return PhaseResult(name=name, episodes_processed=processed, episodes_skipped=0)
+
+    def _run_plan_update_phase(  # noqa: PLR0915
+        self, records: list[EpisodeRecord], staging_dir: Path
+    ) -> PhaseResult:
+        """Re-emit ``plan`` rows at each timestamp the ``interjections`` module produced.
+
+        The ``plan`` module owns the prompt; the ``interjections`` module
+        produced the timestamps. This phase therefore calls back into the
+        ``plan`` module with the interjection timestamps so its existing
+        prompt path is reused.
+        """
+        if not self.plan.enabled or not self.interjections.enabled:
+            return PhaseResult(name="plan_update", episodes_processed=0, episodes_skipped=len(records))
+        processed = 0
+        for record in records:
+            staging = EpisodeStaging(staging_dir, record.episode_index)
+            interjection_rows = [
+                row for row in staging.read("interjections") if row.get("style") == "interjection"
+            ]
+            interjection_times = [float(row["timestamp"]) for row in interjection_rows]
+            interjection_texts = [str(row.get("content") or "") for row in interjection_rows]
+            if interjection_times:
+                self.plan.run_plan_updates(record, staging, interjection_times, interjection_texts)
+                processed += 1
+        # Episodes without any interjections are skipped (no plan refresh
+        # needed); count them so the summary's processed+skipped == total.
+        return PhaseResult(
+            name="plan_update",
+            episodes_processed=processed,
+            episodes_skipped=len(records) - processed,
+        )

src/lerobot/annotations/steerable_pipeline/frames.py

@@ -0,0 +1,494 @@
+#!/usr/bin/env python
+
+# Copyright 2026 The HuggingFace Inc. team. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""Keyframe extraction for the annotation pipeline.
+
+Modules attach decoded camera frames to their VLM prompts so the model can
+ground subtask decomposition, interjection scenarios, and VQA in actual
+visual content. The pipeline shares one provider across modules and one
+episode at a time, with a small per-episode cache so multiple modules
+querying the same timestamp pay decode cost once.
+"""
+
+from __future__ import annotations
+
+import logging
+import threading
+from dataclasses import dataclass, field
+from pathlib import Path
+from typing import Any, Protocol
+
+import PIL.Image
+import torch
+
+from lerobot.datasets.video_utils import decode_video_frames
+
+from .reader import EpisodeRecord
+
+logger = logging.getLogger(__name__)
+
+
+class FrameProvider(Protocol):
+    """Decodes camera frames at episode-relative timestamps."""
+
+    @property
+    def camera_keys(self) -> list[str]:
+        """All ``observation.images.*`` feature keys this provider can decode."""
+
+    def frames_at(
+        self,
+        record: EpisodeRecord,
+        timestamps: list[float],
+        camera_key: str | None = None,
+    ) -> list[Any]:
+        """Return one decoded frame per timestamp from ``camera_key`` (or default).
+
+        Frames are ``torch.Tensor`` (``C, H, W`` uint8) — the shape
+        :func:`lerobot.datasets.video_utils.decode_video_frames` returns.
+        :func:`to_image_blocks` converts them to PIL only at the VLM-message
+        boundary.
+
+        Empty list if the camera is unavailable. ``camera_key=None`` falls back
+        to the provider's default camera so existing single-camera callers
+        (the ``plan`` and ``interjections`` modules) keep working unchanged.
+        """
+
+    def video_for_episode(
+        self,
+        record: EpisodeRecord,
+        max_frames: int,
+        camera_key: str | None = None,
+    ) -> list[Any]:
+        """Return up to ``max_frames`` decoded frames covering the whole episode.
+
+        Sampling is uniform across the episode duration. Frames are
+        ``torch.Tensor`` (``C, H, W`` uint8); :func:`to_video_block` wraps
+        them into one ``{"type":"video", "video":<list>}`` block for a
+        Qwen-VL-compatible model that pools temporally itself. Empty list if
+        no camera available.
+        """
+
+
+@dataclass
+class _NullProvider:
+    """No-op provider used when the dataset has no video keys or in tests."""
+
+    @property
+    def camera_keys(self) -> list[str]:
+        return []
+
+    def frames_at(
+        self,
+        record: EpisodeRecord,
+        timestamps: list[float],
+        camera_key: str | None = None,
+    ) -> list[Any]:
+        return []
+
+    def video_for_episode(
+        self,
+        record: EpisodeRecord,
+        max_frames: int,
+        camera_key: str | None = None,
+    ) -> list[Any]:
+        return []
+
+
+def null_provider() -> FrameProvider:
+    return _NullProvider()
+
+
+@dataclass
+class VideoFrameProvider:
+    """Decodes frames from the dataset's ``observation.images.*`` streams.
+
+    By default the *first* camera key is used for the ``plan`` module
+    (subtask decomposition) and the ``interjections`` module (interjection
+    scenarios) — those prompts care about *what is happening*, not which
+    angle. The ``vqa`` module instead iterates over every camera in
+    :attr:`camera_keys` so each frame's
+    grounded answer (bbox/keypoint/...) is tagged with the camera it was
+    grounded against.
+
+    ``camera_key`` overrides the default-camera choice but does not restrict
+    :attr:`camera_keys`. Pass ``camera_key`` explicitly to ``frames_at`` /
+    ``video_for_episode`` to read a non-default stream.
+
+    Caches up to ``cache_size`` decoded frames per process to keep
+    co-timestamped ``interjections`` + ``plan`` plan-update calls cheap.
+    """
+
+    root: Path
+    camera_key: str | None = None
+    tolerance_s: float = 1e-2
+    cache_size: int = 256
+    # Keyframe decode backend. ``None`` uses the ffmpeg CLI — the
+    # concurrency- and crash-safe default for the pipeline's threaded
+    # decode. Set to ``"torchcodec"`` or ``"pyav"`` to pin an in-process
+    # decoder when the build is known thread-safe.
+    video_backend: str | None = None
+    _meta: Any = field(default=None, init=False, repr=False)
+    _cache: dict = field(default_factory=dict, init=False, repr=False)
+    _camera_keys: list[str] = field(default_factory=list, init=False, repr=False)
+    # Pipeline runs the three module phases under a ThreadPoolExecutor (see
+    # ``ExecutorConfig.episode_parallelism``); guard the dict cache and the
+    # one-shot warn flag against concurrent updates from worker threads.
+    _lock: threading.Lock = field(default_factory=threading.Lock, init=False, repr=False)
+
+    def __post_init__(self) -> None:
+        from lerobot.datasets.dataset_metadata import LeRobotDatasetMetadata  # noqa: PLC0415
+
+        self._meta = LeRobotDatasetMetadata(repo_id="local", root=self.root)
+        # Only ``video_keys`` are decodable here: the clip/decode paths read
+        # ``videos/<key>/from_timestamp`` from episode metadata, which exists
+        # only for video-stored cameras. Image-stored cameras (also in
+        # ``camera_keys``) would KeyError, so restrict the list — and the
+        # default — to video keys.
+        keys = list(self._meta.video_keys)
+        # Last-resort fallback: if metadata didn't surface any video keys but
+        # the caller explicitly named a camera (``--vlm.camera_key=...``),
+        # trust them — the key is by definition known to exist on the dataset.
+        if not keys and self.camera_key:
+            keys = [self.camera_key]
+        self._camera_keys = keys
+        if self.camera_key is None:
+            self.camera_key = keys[0] if keys else None
+
+    @property
+    def camera_keys(self) -> list[str]:
+        """All ``observation.images.*`` keys available on this dataset."""
+        return list(self._camera_keys)
+
+    def frames_at(
+        self,
+        record: EpisodeRecord,
+        timestamps: list[float],
+        camera_key: str | None = None,
+    ) -> list[Any]:
+        target = camera_key if camera_key is not None else self.camera_key
+        if not timestamps or target is None:
+            return []
+
+        out: list[Any] = []
+        misses: list[float] = []
+        miss_indices: list[int] = []
+        with self._lock:
+            for i, ts in enumerate(timestamps):
+                key = (record.episode_index, target, round(float(ts), 6))
+                cached = self._cache.get(key)
+                if cached is not None:
+                    out.append(cached)
+                else:
+                    out.append(None)
+                    misses.append(float(ts))
+                    miss_indices.append(i)
+
+        if misses:
+            decoded = self._decode(record.episode_index, misses, target)
+            # ``_decode`` returns exactly one frame per requested timestamp,
+            # or an empty list if decoding failed wholesale. A partial list
+            # would mean a frame/timestamp misalignment, so only pair them up
+            # when the counts match (``strict=True`` then guards regressions).
+            if len(decoded) == len(miss_indices):
+                with self._lock:
+                    for i, frame in zip(miss_indices, decoded, strict=True):
+                        out[i] = frame
+                        key = (record.episode_index, target, round(float(timestamps[i]), 6))
+                        if len(self._cache) >= self.cache_size:
+                            self._cache.pop(next(iter(self._cache)))
+                        self._cache[key] = frame
+        # filter out any None left over from decode failures
+        return [frame for frame in out if frame is not None]
+
+    def video_for_episode(
+        self,
+        record: EpisodeRecord,
+        max_frames: int,
+        camera_key: str | None = None,
+    ) -> list[Any]:
+        """Return up to ``max_frames`` frames uniformly sampled across the episode.
+
+        The whole episode duration is covered; the model picks subtask
+        boundaries from the temporal pooling it does internally. Frames are
+        ``torch.Tensor`` (see :meth:`frames_at`).
+        """
+        target = camera_key if camera_key is not None else self.camera_key
+        if max_frames <= 0 or target is None or not record.frame_timestamps:
+            return []
+        n_frames = min(max_frames, len(record.frame_timestamps))
+        if n_frames == len(record.frame_timestamps):
+            timestamps = list(record.frame_timestamps)
+        else:
+            t0 = record.frame_timestamps[0]
+            t_last = record.frame_timestamps[-1]
+            if t_last <= t0:
+                timestamps = [float(t0)] * n_frames
+            else:
+                step = (t_last - t0) / (n_frames - 1) if n_frames > 1 else 0.0
+                timestamps = [float(t0 + i * step) for i in range(n_frames)]
+        return self.frames_at(record, timestamps, camera_key=target)
+
+    def episode_clip_path(self, record: EpisodeRecord, cache_dir: Path) -> Path | None:
+        """Extract the episode's subclip to ``cache_dir/ep_{idx:06d}.mp4``.
+
+        Returns ``None`` if the dataset has no video tracks. Skips
+        re-extract when the cached clip already exists. Re-encodes to
+        H.264 (libx264) so the resulting mp4 is decodable by every
+        downstream video processor — stream-copy would inherit the
+        source codec (often AV1 in modern LeRobot datasets), which
+        vllm's libav build cannot decode.
+        """
+        import subprocess  # noqa: PLC0415
+
+        if self.camera_key is None:
+            return None
+        cache_dir.mkdir(parents=True, exist_ok=True)
+        out_path = cache_dir / f"ep_{record.episode_index:06d}.mp4"
+        if out_path.exists() and out_path.stat().st_size > 0:
+            return out_path
+        ep = self._meta.episodes[record.episode_index]
+        from_timestamp = float(ep[f"videos/{self.camera_key}/from_timestamp"])
+        to_timestamp = float(ep[f"videos/{self.camera_key}/to_timestamp"])
+        src = self.root / self._meta.get_video_file_path(record.episode_index, self.camera_key)
+        cmd = [
+            "ffmpeg",
+            "-y",
+            "-loglevel",
+            "error",
+            "-ss",
+            f"{from_timestamp:.3f}",
+            "-to",
+            f"{to_timestamp:.3f}",
+            "-i",
+            str(src),
+            "-c:v",
+            "libx264",
+            "-preset",
+            "ultrafast",
+            "-crf",
+            "23",
+            "-pix_fmt",
+            "yuv420p",
+            "-an",
+            str(out_path),
+        ]
+        try:
+            subprocess.run(cmd, check=True, timeout=300)
+        except (subprocess.CalledProcessError, subprocess.TimeoutExpired, FileNotFoundError):
+            return None
+        return out_path if out_path.exists() and out_path.stat().st_size > 0 else None
+
+    def _decode(self, episode_index: int, timestamps: list[float], camera_key: str) -> list[Any]:
+        """Decode ``timestamps`` from the episode's video as ``(C, H, W)`` tensors.
+
+        Delegates to :func:`lerobot.datasets.video_utils.decode_video_frames`
+        (torchcodec by default, PyAV fallback) rather than a bespoke decoder.
+        Returns one frame per requested timestamp, or ``[]`` if decoding
+        failed wholesale — callers treat ``[]`` as "no frames available".
+        """
+        ep = self._meta.episodes[episode_index]
+        from_timestamp = ep[f"videos/{camera_key}/from_timestamp"]
+        shifted = [from_timestamp + ts for ts in timestamps]
+        video_path = self.root / self._meta.get_video_file_path(episode_index, camera_key)
+
+        # Default to the ffmpeg CLI. The pipeline decodes under a 16-wide
+        # ThreadPoolExecutor and the in-process decoders are unsafe there:
+        # torchcodec is not thread-safe and SIGSEGVs under concurrent decode
+        # (a crash no try/except can catch), PyAV can likewise segfault on
+        # AV1, and lerobot's ``pyav`` backend routes through the removed
+        # ``torchvision.io.VideoReader``. ``_decode_frames_ffmpeg`` shells
+        # out per frame: each decode is an isolated child process, so it is
+        # both crash-safe and concurrency-safe. ``video_backend`` can pin
+        # ``torchcodec`` / ``pyav`` explicitly for callers that know their
+        # build is safe.
+        chain = [self.video_backend] if self.video_backend else ["ffmpeg"]
+
+        exc: Exception | None = None
+        for backend in chain:
+            try:
+                if backend == "ffmpeg":
+                    return _decode_frames_ffmpeg(video_path, shifted)
+                if backend in ("pyav", "av"):
+                    return _decode_frames_av(video_path, shifted)
+                # Stacked ``(N, C, H, W)`` uint8 tensor; one row per timestamp.
+                decoded = decode_video_frames(
+                    video_path, shifted, self.tolerance_s, backend=backend, return_uint8=True
+                )
+                return list(decoded)
+            except Exception as e:  # noqa: PERF203
+                exc = e
+
+        # Every backend raised. Log loudly the first time so a silent
+        # vqa-module no-op (every prompt skipped because frames_at returned
+        # []) is debuggable from the job log instead of post-hoc parquet
+        # inspection. Subsequent failures stay quiet.
+        with self._lock:
+            already_warned = getattr(self, "_warned_decode_fail", False)
+            if not already_warned:
+                self._warned_decode_fail = True
+        if not already_warned:
+            logger.warning(
+                "VideoFrameProvider._decode failed for episode=%s camera=%s video_path=%s backends=%s: %s",
+                episode_index,
+                camera_key,
+                video_path,
+                chain,
+                exc,
+                exc_info=exc,
+            )
+        return []
+
+
+def make_frame_provider(
+    root: Path, camera_key: str | None = None, video_backend: str | None = None
+) -> FrameProvider:
+    """Build a :class:`VideoFrameProvider` if videos are present, else null."""
+    try:
+        provider = VideoFrameProvider(root=root, camera_key=camera_key, video_backend=video_backend)
+    except Exception:
+        return null_provider()
+    if provider.camera_key is None:
+        return null_provider()
+    return provider
+
+
+def _decode_frames_ffmpeg(video_path: Path, timestamps: list[float]) -> list[Any]:
+    """Decode the frames nearest to ``timestamps`` via the ffmpeg CLI.
+
+    Runs one ``ffmpeg`` process per timestamp, seeking with ``-ss`` and
+    piping a single PNG to stdout. Unlike the in-process decoders this
+    survives a hostile container: a full ffmpeg build decodes AV1 (the codec
+    modern LeRobot datasets use) where torchcodec raises and PyAV can
+    SIGSEGV, and a crash stays isolated to the child process — a non-zero
+    exit is a catchable error, not a segfault of the whole job. Returns one
+    ``(C, H, W)`` uint8 tensor per timestamp.
+    """
+    import io  # noqa: PLC0415
+    import subprocess  # noqa: PLC0415
+
+    import numpy as np  # noqa: PLC0415
+
+    frames: list[Any] = []
+    for ts in timestamps:
+        proc = subprocess.run(
+            [
+                "ffmpeg",
+                "-nostdin",
+                "-loglevel",
+                "error",
+                "-ss",
+                f"{max(ts, 0.0):.3f}",
+                "-i",
+                str(video_path),
+                "-frames:v",
+                "1",
+                "-f",
+                "image2pipe",
+                "-vcodec",
+                "png",
+                "pipe:1",
+            ],
+            capture_output=True,
+            check=True,
+            timeout=120,
+        )
+        if not proc.stdout:
+            raise RuntimeError(f"ffmpeg returned no frame for t={ts:.3f}s of {video_path}")
+        img = PIL.Image.open(io.BytesIO(proc.stdout)).convert("RGB")
+        frames.append(torch.from_numpy(np.asarray(img).copy()).permute(2, 0, 1).contiguous())
+    return frames
+
+
+def _decode_frames_av(video_path: Path, timestamps: list[float]) -> list[Any]:
+    """Decode the frames nearest to ``timestamps`` using PyAV directly.
+
+    lerobot's ``decode_video_frames(backend="pyav")`` routes through
+    ``torchvision.io.VideoReader``, removed in torchvision 0.23+. This helper
+    talks to the ``av`` package directly. Note PyAV can SIGSEGV on AV1
+    streams in some builds — prefer ``_decode_frames_ffmpeg`` as the default
+    fallback; this stays available behind ``video_backend="pyav"``. Returns
+    one ``(C, H, W)`` uint8 tensor per timestamp.
+    """
+    import av  # noqa: PLC0415
+
+    first_ts = min(timestamps)
+    last_ts = max(timestamps)
+    loaded_frames: list[torch.Tensor] = []
+    loaded_ts: list[float] = []
+    with av.open(str(video_path)) as container:
+        stream = container.streams.video[0]
+        # Seek to the keyframe at or before the first requested timestamp.
+        offset = max(int(first_ts / stream.time_base), 0) if stream.time_base else 0
+        container.seek(offset, stream=stream, backward=True, any_frame=False)
+        for idx, frame in enumerate(container.decode(stream)):
+            ts = frame.time
+            if ts is None:
+                ts = float(frame.pts * stream.time_base) if frame.pts is not None else float(idx)
+            loaded_ts.append(ts)
+            loaded_frames.append(
+                torch.from_numpy(frame.to_ndarray(format="rgb24")).permute(2, 0, 1).contiguous()
+            )
+            if ts >= last_ts:
+                break
+    if not loaded_frames:
+        raise RuntimeError(f"PyAV decoded no frames from {video_path}")
+    ts_tensor = torch.tensor(loaded_ts)
+    return [loaded_frames[int(torch.argmin((ts_tensor - q).abs()))] for q in timestamps]
+
+
+def _frame_to_pil(frame: Any) -> Any:
+    """Materialise a decoded frame as a ``PIL.Image`` for the VLM message.
+
+    Frames flow through the provider as ``torch.Tensor`` (``C, H, W`` uint8,
+    straight from :func:`decode_video_frames`); PIL is only created here, at
+    the VLM-message boundary, because the chat backends expect PIL images /
+    data URLs. Non-tensor inputs (e.g. test stubs) pass through untouched.
+    """
+    if not isinstance(frame, torch.Tensor):
+        return frame
+    array = frame.detach().cpu()
+    if array.ndim == 3 and array.shape[0] in (1, 3):
+        array = array.permute(1, 2, 0)  # (C, H, W) -> (H, W, C)
+    if array.shape[-1] == 1:
+        array = array.squeeze(-1)
+    return PIL.Image.fromarray(array.to(torch.uint8).numpy())
+
+
+def to_image_blocks(frames: list[Any]) -> list[dict[str, Any]]:
+    """Convert decoded frames to Qwen-VL-compatible image content blocks."""
+    return [{"type": "image", "image": _frame_to_pil(frame)} for frame in frames]
+
+
+def to_video_block(frames: list[Any]) -> list[dict[str, Any]]:
+    """Wrap a list of decoded frames as one Qwen-VL video block.
+
+    Returns ``[]`` when the list is empty, so the caller can splat the result
+    into a content array without a separate emptiness check.
+    """
+    if not frames:
+        return []
+    return [{"type": "video", "video": [_frame_to_pil(frame) for frame in frames]}]
+
+
+def to_video_url_block(url: str | None, fps: float = 2.0) -> list[dict[str, Any]]:
+    """Wrap a video file URL as one ``video_url`` block.
+
+    Used by the ``openai`` backend (transformers serve / vllm serve /
+    ktransformers serve), where the server handles frame sampling.
+    Returns ``[]`` when ``url`` is ``None`` so the caller can splat.
+    """
+    if not url:
+        return []
+    return [{"type": "video_url", "video_url": {"url": url}, "fps": fps}]

src/lerobot/annotations/steerable_pipeline/modules/__init__.py

@@ -0,0 +1,25 @@
+#!/usr/bin/env python
+
+# Copyright 2026 The HuggingFace Inc. team. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from .general_vqa import GeneralVqaModule
+from .interjections_and_speech import InterjectionsAndSpeechModule
+from .plan_subtasks_memory import PlanSubtasksMemoryModule
+
+__all__ = [
+    "GeneralVqaModule",
+    "InterjectionsAndSpeechModule",
+    "PlanSubtasksMemoryModule",
+]

src/lerobot/annotations/steerable_pipeline/modules/general_vqa.py

@@ -0,0 +1,238 @@
+#!/usr/bin/env python
+
+# Copyright 2026 The HuggingFace Inc. team. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""``vqa`` module: general VQA at a timed cadence.
+
+Every ``1/hz`` seconds an emission tick fires; each tick anchors ``K``
+consecutive frames, and every anchored frame gets its own VQA pair. Each
+pair is grounded on that single anchor frame — there is no per-pair frame
+window. For datasets with multiple cameras, every anchored frame produces
+one ``(vqa, user)`` + ``(vqa, assistant)`` pair *per camera*: each pair is
+generated against that camera's frame and stamped with the matching
+``camera`` field on the emitted rows. The resolver disambiguates via
+``camera=...``; recipes that consume VQA do so through one sub-recipe
+per camera (see ``recipes/pi05_hirobot.yaml``).
+
+Within a single (frame, camera) we still emit at most one ``(vqa, user)``
+and one ``(vqa, assistant)`` row, so the resolver contract stays scalar.
+
+Question types covered (per the plan's ``vqa`` table): bbox, keypoint,
+count, attribute, spatial. The assistant's ``content`` is a JSON string
+whose schema depends on the question type. Malformed JSON triggers one
+retry inside :meth:`VlmClient.generate_json`.
+"""
+
+from __future__ import annotations
+
+import json
+import logging
+import random
+from collections.abc import Sequence
+from dataclasses import dataclass, field
+from typing import Any
+
+from ..config import VqaConfig
+from ..frames import FrameProvider, null_provider, to_image_blocks
+from ..prompts import load as load_prompt
+from ..reader import EpisodeRecord
+from ..staging import EpisodeStaging
+from ..validator import classify_vqa_answer
+from ..vlm_client import VlmClient
+
+
+def _emission_anchor_indices(frame_timestamps: Sequence[float], hz: float, k: int) -> list[int]:
+    """Return the relative frame indices to anchor VQA emissions to.
+
+    For each emission tick (every ``1/hz`` seconds), we anchor ``k``
+    consecutive frames starting at the tick. Ticks fall on the nearest
+    available source frame timestamp.
+    """
+    if hz <= 0 or k <= 0 or not frame_timestamps:
+        return []
+    t0 = frame_timestamps[0]
+    t_last = frame_timestamps[-1]
+    period = 1.0 / hz
+    indices: list[int] = []
+    t = t0
+    while t <= t_last + 1e-9:
+        # find the index of the nearest frame to t
+        nearest_i = min(range(len(frame_timestamps)), key=lambda i: abs(frame_timestamps[i] - t))
+        for offset in range(k):
+            j = nearest_i + offset
+            if j >= len(frame_timestamps):
+                break
+            if not indices or indices[-1] != j:
+                indices.append(j)
+        t += period
+    # dedupe while preserving order
+    seen: set[int] = set()
+    deduped: list[int] = []
+    for i in indices:
+        if i in seen:
+            continue
+        seen.add(i)
+        deduped.append(i)
+    return deduped
+
+
+@dataclass
+class GeneralVqaModule:
+    """Emit grounded VQA pairs at a timed cadence."""
+
+    vlm: VlmClient
+    config: VqaConfig
+    seed: int = 1729
+    frame_provider: FrameProvider = field(default_factory=null_provider)
+
+    @property
+    def enabled(self) -> bool:
+        return self.config.enabled
+
+    def run_episode(self, record: EpisodeRecord, staging: EpisodeStaging) -> None:
+        if not record.frame_timestamps:
+            staging.write("vqa", [])
+            return
+        rng = random.Random(f"{self.seed}:{record.episode_index}:vqa")
+        anchor_idx = _emission_anchor_indices(
+            record.frame_timestamps, self.config.vqa_emission_hz, self.config.K
+        )
+        cameras = self._target_cameras()
+        if not cameras:
+            # No camera available — emit nothing rather than producing
+            # untagged rows that would fail validation. Surface a loud one-
+            # time warning so this is never silently a no-op.
+            if not getattr(self, "_warned_no_camera", False):
+                logging.getLogger(__name__).warning(
+                    "vqa module found no cameras on the frame provider — "
+                    "every episode will emit zero VQA rows. Check that the "
+                    "dataset declares observation.images.* features in "
+                    "meta/info.json; passing --vlm.camera_key=<key> at the "
+                    "CLI now also seeds the cameras list as a fallback."
+                )
+                self._warned_no_camera = True
+            staging.write("vqa", [])
+            return
+
+        # Build all messages first (one per (frame, camera)), then issue them
+        # as a single batched generate_json call so the client can fan them
+        # out concurrently.
+        per_call: list[tuple[float, str, str, list[dict[str, Any]]]] = []
+        for idx in anchor_idx:
+            ts = float(record.frame_timestamps[idx])
+            qtype = rng.choice(self.config.question_types)
+            for camera in cameras:
+                messages = self._build_messages(record, qtype, ts, camera)
+                # Skip cameras that decoded to zero frames at this ts: no point
+                # asking the VLM to ground a bbox without an image.
+                if not _has_image_block(messages):
+                    continue
+                per_call.append((ts, camera, qtype, messages))
+
+        if not per_call:
+            staging.write("vqa", [])
+            return
+
+        results = self.vlm.generate_json([m for _, _, _, m in per_call])
+
+        rows: list[dict[str, Any]] = []
+        for (ts, camera, _qtype, _messages), result in zip(per_call, results, strict=True):
+            qa = self._postprocess(result)
+            if qa is None:
+                continue
+            question, answer = qa
+            rows.append(
+                {
+                    "role": "user",
+                    "content": question,
+                    "style": "vqa",
+                    "timestamp": ts,
+                    "camera": camera,
+                    "tool_calls": None,
+                }
+            )
+            rows.append(
+                {
+                    "role": "assistant",
+                    "content": json.dumps(answer, sort_keys=True),
+                    "style": "vqa",
+                    "timestamp": ts,
+                    "camera": camera,
+                    "tool_calls": None,
+                }
+            )
+        staging.write("vqa", rows)
+
+    def _target_cameras(self) -> list[str]:
+        """Return the cameras the ``vqa`` module should iterate per anchored frame.
+
+        Defaults to every camera the provider exposes. Datasets with no
+        cameras (or test/null providers) yield an empty list, which makes
+        ``run_episode`` a no-op.
+
+        When ``config.restrict_to_default_camera`` is set, VQA grounds on
+        only the provider's default camera (the single ``--vlm.camera_key``
+        stream), matching the plan / interjection modules so the whole
+        pipeline focuses on one view.
+        """
+        all_cameras = list(getattr(self.frame_provider, "camera_keys", []) or [])
+        if getattr(self.config, "restrict_to_default_camera", False):
+            default = getattr(self.frame_provider, "camera_key", None)
+            if default and default in all_cameras:
+                return [default]
+            if default:
+                return [default]
+        return all_cameras
+
+    def _build_messages(
+        self,
+        record: EpisodeRecord,
+        question_type: str,
+        frame_timestamp: float,
+        camera_key: str,
+    ) -> list[dict[str, Any]]:
+        prompt = load_prompt("module_3_vqa").format(
+            episode_task=record.episode_task,
+            question_type=question_type,
+        )
+        images = self.frame_provider.frames_at(record, [frame_timestamp], camera_key=camera_key)
+        content = [*to_image_blocks(images), {"type": "text", "text": prompt}]
+        return [{"role": "user", "content": content}]
+
+    def _postprocess(self, result: Any) -> tuple[str, dict[str, Any]] | None:
+        if not isinstance(result, dict):
+            return None
+        question = result.get("question")
+        answer = result.get("answer")
+        if not isinstance(question, str) or not question.strip():
+            return None
+        if not isinstance(answer, dict):
+            return None
+        # The validator will enforce shape; here we just sanity-check that the
+        # answer matches *some* known shape so we can drop garbage early.
+        if classify_vqa_answer(answer) is None:
+            return None
+        return question.strip(), answer
+
+
+def _has_image_block(messages: list[dict[str, Any]]) -> bool:
+    """Return True if any user content block is a populated image block."""
+    for msg in messages:
+        content = msg.get("content")
+        if not isinstance(content, list):
+            continue
+        for block in content:
+            if isinstance(block, dict) and block.get("type") == "image":
+                return True
+    return False

src/lerobot/annotations/steerable_pipeline/modules/interjections_and_speech.py

@@ -0,0 +1,210 @@
+#!/usr/bin/env python
+
+# Copyright 2026 The HuggingFace Inc. team. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""``interjections`` module: interjections + paired speech (EVENT styles + speech atoms).
+
+Two sub-passes:
+
+1. At ``t=0``, emit ONLY a speech tool-call atom (acknowledgement of the
+   canonical task). No interjection row — the canonical task is already the
+   user utterance from ``meta/tasks.parquet``.
+
+2. For mid-episode interruptions, emit a co-timestamped pair:
+       {role:user, style:interjection, content:<text>}
+       speech atom (role:assistant, style:None, tool_calls=[say(...)])
+   Both rows go in ``language_events`` at the same timestamp.
+
+The ``plan`` module's :meth:`run_plan_updates` reuses this module's
+interjection timestamps to refresh the ``plan`` row at the same instant.
+"""
+
+from __future__ import annotations
+
+import random
+from collections.abc import Sequence
+from dataclasses import dataclass, field
+from typing import Any
+
+from ..config import InterjectionsConfig
+from ..frames import FrameProvider, null_provider, to_image_blocks
+from ..prompts import load as load_prompt
+from ..reader import EpisodeRecord, reconstruct_subtask_spans, snap_to_frame
+from ..staging import EpisodeStaging
+from ..vlm_client import VlmClient
+from ..writer import speech_atom
+
+
+@dataclass
+class InterjectionsAndSpeechModule:
+    """Generate task-start speech and mid-episode interjection/speech pairs."""
+
+    vlm: VlmClient
+    config: InterjectionsConfig
+    seed: int = 1729
+    frame_provider: FrameProvider = field(default_factory=null_provider)
+
+    @property
+    def enabled(self) -> bool:
+        return self.config.enabled
+
+    def run_episode(self, record: EpisodeRecord, staging: EpisodeStaging) -> None:
+        rows: list[dict[str, Any]] = []
+        if record.frame_timestamps:
+            t0 = float(record.frame_timestamps[0])
+            initial = self._initial_speech(record)
+            if initial:
+                rows.append(speech_atom(t0, initial))
+        # Pull the ``plan`` module's subtask spans for this episode so the
+        # interjection prompt can ground itself in the actual current
+        # subtask at each chosen timestamp. The ``plan`` module ran first.
+        episode_end_t = float(record.frame_timestamps[-1]) if record.frame_timestamps else None
+        subtask_spans = reconstruct_subtask_spans(staging.read("plan"), episode_end_t=episode_end_t)
+        rows.extend(self._mid_episode_interjections(record, subtask_spans))
+        staging.write("interjections", rows)
+
+    @staticmethod
+    def _subtask_at(spans: Sequence[dict[str, Any]], t: float) -> str | None:
+        current: str | None = None
+        for span in spans:
+            if float(span["start"]) <= t:
+                current = span.get("text")
+            else:
+                break
+        return current
+
+    def _initial_speech(self, record: EpisodeRecord) -> str | None:
+        prompt = load_prompt("module_2_initial_speech").format(
+            episode_task=record.episode_task,
+        )
+        messages = [{"role": "user", "content": [{"type": "text", "text": prompt}]}]
+        result = self.vlm.generate_json([messages])[0]
+        if isinstance(result, dict) and isinstance(result.get("text"), str):
+            text = result["text"].strip()
+            if text:
+                return text
+        return None
+
+    def _mid_episode_interjections(
+        self,
+        record: EpisodeRecord,
+        subtask_spans: Sequence[dict[str, Any]],
+    ) -> list[dict[str, Any]]:
+        """Generate interjections aligned with the actual demo trajectory.
+
+        Teleop data is frozen — the robot already executed every step in
+        the video. A *counterfactual* interjection like "actually skip
+        the wipe" contradicts what then happens in the video, which is
+        what qwen36moe-10/11 surfaced as low-quality interjections.
+
+        Instead, anchor every interjection at a subtask boundary and
+        write it as a natural user request for the *upcoming* subtask.
+        The robot's visible next behavior IS the interjection's effect,
+        so the training signal stays consistent: interjection text →
+        plan refresh → action stream all line up.
+        """
+        if self.config.max_interjections_per_episode <= 0:
+            return []
+        if len(subtask_spans) < 2:
+            # Need at least one transition (subtask 0 → subtask 1).
+            return []
+        # Deterministic per-episode RNG so reruns are stable across SLURM jobs.
+        rng = random.Random(f"{self.seed}:{record.episode_index}:interjection")
+
+        # Boundaries: the start time of every subtask except the first
+        # (which is just t0 and is covered by the initial-task speech atom).
+        boundaries: list[tuple[float, str, str]] = []
+        for i in range(1, len(subtask_spans)):
+            ts = float(subtask_spans[i]["start"])
+            if ts < self.config.interjection_min_t:
+                continue
+            prev_text = (subtask_spans[i - 1].get("text") or "").strip()
+            next_text = (subtask_spans[i].get("text") or "").strip()
+            if not next_text:
+                continue
+            boundaries.append((ts, prev_text, next_text))
+        if not boundaries:
+            return []
+
+        n = min(self.config.max_interjections_per_episode, len(boundaries))
+        chosen = sorted(rng.sample(boundaries, n), key=lambda b: b[0])
+
+        out: list[dict[str, Any]] = []
+        for t, prev_subtask, next_subtask in chosen:
+            t_snap = snap_to_frame(t, record.frame_timestamps)
+            # Window straddles the boundary so the VLM sees the end of the
+            # previous subtask and the start of the next one — same
+            # conditioning the policy will see at training time.
+            window_ts = self._window_timestamps(t_snap, record.frame_timestamps)
+            prompt = load_prompt("module_2_interjection").format(
+                episode_task=record.episode_task,
+                prev_subtask=prev_subtask or "(starting from initial state)",
+                next_subtask=next_subtask,
+                timestamp=t_snap,
+                window_seconds=self.config.interjection_window_seconds,
+            )
+            images = self.frame_provider.frames_at(record, window_ts)
+            content = [*to_image_blocks(images), {"type": "text", "text": prompt}]
+            messages = [{"role": "user", "content": content}]
+            result = self.vlm.generate_json([messages])[0]
+            if not isinstance(result, dict):
+                continue
+            interjection_text = result.get("interjection")
+            speech_text = result.get("speech")
+            if not isinstance(interjection_text, str) or not interjection_text.strip():
+                continue
+            if not isinstance(speech_text, str) or not speech_text.strip():
+                continue
+            out.append(
+                {
+                    "role": "user",
+                    "content": interjection_text.strip(),
+                    "style": "interjection",
+                    "timestamp": t_snap,
+                    "tool_calls": None,
+                }
+            )
+            out.append(speech_atom(t_snap, speech_text.strip()))
+        return out
+
+    def _window_timestamps(self, t_anchor: float, frame_timestamps: Sequence[float]) -> list[float]:
+        """Return a small set of frame timestamps centered on ``t_anchor``.
+
+        The window straddles the subtask boundary the interjection sits
+        on: roughly half the frames cover the end of the previous
+        subtask, half cover the start of the next one. The VLM therefore
+        sees BOTH what just finished AND what's about to start, which is
+        the conditioning we need to write a natural "now please do X"
+        request that matches the visible upcoming behavior.
+        """
+        if not frame_timestamps:
+            return [t_anchor]
+        n = max(1, int(self.config.interjection_window_frames))
+        if n == 1:
+            return [t_anchor]
+        window = float(self.config.interjection_window_seconds)
+        step = window / max(1, n - 1)
+        # Center the window on the anchor so half lands before, half after.
+        start_offset = -window / 2.0
+        targets = [t_anchor + start_offset + step * i for i in range(n)]
+        last_ts = float(frame_timestamps[-1])
+        snapped: list[float] = []
+        seen: set[float] = set()
+        for tgt in targets:
+            clamped = min(last_ts, max(0.0, tgt))
+            t = snap_to_frame(clamped, frame_timestamps)
+            if t not in seen:
+                seen.add(t)
+                snapped.append(t)
+        return snapped or [t_anchor]

src/lerobot/annotations/steerable_pipeline/modules/plan_subtasks_memory.py

@@ -0,0 +1,892 @@
+#!/usr/bin/env python
+
+# Copyright 2026 The HuggingFace Inc. team. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""``plan`` module: subtask decomposition + plan + memory (PERSISTENT styles)."""
+
+from __future__ import annotations
+
+import json
+import logging
+from collections.abc import Sequence
+from dataclasses import dataclass, field
+from pathlib import Path
+from typing import Any
+
+from ..config import PlanConfig
+from ..frames import (
+    FrameProvider,
+    VideoFrameProvider,
+    null_provider,
+    to_image_blocks,
+    to_video_block,
+    to_video_url_block,
+)
+from ..prompts import load as load_prompt
+from ..reader import EpisodeRecord, reconstruct_subtask_spans, snap_to_frame
+from ..staging import EpisodeStaging
+from ..vlm_client import VlmClient
+
+logger = logging.getLogger(__name__)
+
+
+@dataclass
+class PlanSubtasksMemoryModule:
+    """Generate subtask spans, plan, and memory rows.
+
+    All output is persistent (lives in ``language_persistent``):
+
+    - ``subtask`` rows: one per span, stamped at the span's *start* timestamp
+      (snapped to an exact frame).
+    - ``plan`` rows: emitted at ``t=0``; refreshed at every interjection
+      timestamp via :meth:`run_plan_updates` (called by the executor after
+      the ``interjections`` module completes).
+    - ``memory`` rows: emitted at each subtask boundary (= subtask start
+      timestamp from the second subtask onward).
+    """
+
+    vlm: VlmClient
+    config: PlanConfig
+    frame_provider: FrameProvider = field(default_factory=null_provider)
+
+    @property
+    def enabled(self) -> bool:
+        return self.config.enabled
+
+    def run_episode(self, record: EpisodeRecord, staging: EpisodeStaging) -> None:
+        rows: list[dict[str, Any]] = []
+        # Resolve the task that drives every other ``plan``-module prompt.
+        # May be the canonical ``record.episode_task`` (default), or a fresh
+        # description derived from the video when the canonical task is
+        # empty / placeholder / forced-off (see PlanConfig.derive_task_*).
+        effective_task = self._resolve_effective_task(record)
+        # ``task_aug`` rows at t=0 (role=user), one per rephrasing — the
+        # message renderer rotates ``${task}`` deterministically through
+        # them so the policy sees diverse phrasings during training.
+        # Two paths:
+        #   * ``task_aug_axes.enabled=True`` — structured 5-axis taxonomy
+        #     (synonym / omit_arm / omit_orientation / omit_grasp_method
+        #     / combined). Replaces the free-form rephrasings flow.
+        #   * Otherwise — free-form ``n_task_rephrasings`` (original).
+        t0 = float(record.frame_timestamps[0]) if record.frame_timestamps else 0.0
+        axes_cfg = self.config.task_aug_axes
+        if axes_cfg.enabled and effective_task:
+            variants = self._generate_task_aug_by_axes(effective_task, axes_cfg)
+            seen: set[str] = set()
+            ordered = [effective_task, *variants]
+            for phrasing in ordered:
+                key = phrasing.strip()
+                if not key or key in seen:
+                    continue
+                seen.add(key)
+                rows.append(
+                    {
+                        "role": "user",
+                        "content": key,
+                        "style": "task_aug",
+                        "timestamp": t0,
+                        "tool_calls": None,
+                    }
+                )
+        elif self.config.n_task_rephrasings > 0 and effective_task:
+            rephrasings = self._generate_task_rephrasings(effective_task, n=self.config.n_task_rephrasings)
+            # Always include the effective task itself as the first variant
+            # so the rotation is guaranteed to cover the source-of-truth
+            # phrasing, not just synthetic alternatives.
+            seen = set()
+            ordered = [effective_task, *rephrasings]
+            for phrasing in ordered:
+                key = phrasing.strip()
+                if not key or key in seen:
+                    continue
+                seen.add(key)
+                rows.append(
+                    {
+                        "role": "user",
+                        "content": key,
+                        "style": "task_aug",
+                        "timestamp": t0,
+                        "tool_calls": None,
+                    }
+                )
+
+        subtask_spans = self._generate_subtasks(record, task=effective_task)
+
+        # ----------------------------------------------------------------
+        # Phase 1a: structured per-subtask action records (additive)
+        # ----------------------------------------------------------------
+        # When enabled, for every subtask span we ask the VLM for a typed
+        # ActionRecord (verb / object / arm / grasp_type / destination /
+        # mistake) and emit it as a separate ``style="action_record"``
+        # row for downstream use. This is purely additive — it never
+        # touches the VLM's subtask text (reconstructing subtask text
+        # from these fields was too easy to hallucinate on tasks that
+        # don't fit the manipulation schema).
+        records_cfg = self.config.action_records
+        action_records: list[dict[str, Any] | None] = [None] * len(subtask_spans)
+        if records_cfg.enabled and subtask_spans:
+            for i, span in enumerate(subtask_spans):
+                rec = self._extract_action_record(record, span, effective_task)
+                if rec is not None:
+                    action_records[i] = rec
+
+        # subtask rows
+        for i, span in enumerate(subtask_spans):
+            rows.append(
+                {
+                    "role": "assistant",
+                    "content": span["text"],
+                    "style": "subtask",
+                    "timestamp": snap_to_frame(span["start"], record.frame_timestamps),
+                    "tool_calls": None,
+                }
+            )
+            if records_cfg.enabled and records_cfg.emit_record_row and action_records[i] is not None:
+                rows.append(
+                    {
+                        "role": "assistant",
+                        "content": json.dumps(action_records[i], sort_keys=True),
+                        "style": "action_record",
+                        "timestamp": snap_to_frame(span["start"], record.frame_timestamps),
+                        "tool_calls": None,
+                    }
+                )
+        # Plan rows at every subtask boundary — including t=0 (start of
+        # the first subtask). Because the plan is just a numbered list
+        # of *still-todo* subtasks, re-emitting at each boundary makes
+        # the active plan shrink as work progresses: at frame t the
+        # rendered ``${plan}`` is the most recent emission, which
+        # contains exactly the subtasks that started at or after the
+        # current span. Saves the runtime from having to derive
+        # "what's still left" at inference time.
+        if self.config.emit_plan:
+            for span in subtask_spans:
+                boundary_t = snap_to_frame(span["start"], record.frame_timestamps)
+                plan_text = self._generate_plan(
+                    record, subtask_spans, refresh_t=boundary_t, task=effective_task
+                )
+                if plan_text is not None:
+                    rows.append(
+                        {
+                            "role": "assistant",
+                            "content": plan_text,
+                            "style": "plan",
+                            "timestamp": float(boundary_t),
+                            "tool_calls": None,
+                        }
+                    )
+        # memory rows at every subtask boundary except the very first start
+        prior_memory = ""
+        for i, span in enumerate(subtask_spans[1:], start=1):
+            completed = subtask_spans[i - 1]["text"]
+            remaining = [s["text"] for s in subtask_spans[i:]]
+            mem_text = self._generate_memory(record, prior_memory, completed, remaining, task=effective_task)
+            if mem_text:
+                ts = snap_to_frame(span["start"], record.frame_timestamps)
+                rows.append(
+                    {
+                        "role": "assistant",
+                        "content": mem_text,
+                        "style": "memory",
+                        "timestamp": ts,
+                        "tool_calls": None,
+                    }
+                )
+                prior_memory = mem_text
+        staging.write("plan", rows)
+
+    # ------------------------------------------------------------------
+    # Task derivation + rephrasings
+    # ------------------------------------------------------------------
+
+    _PLACEHOLDER_TASKS: frozenset[str] = frozenset(
+        {
+            "debug",
+            "test",
+            "tbd",
+            "todo",
+            "n/a",
+            "na",
+            "untitled",
+            "unnamed",
+            "default",
+            "placeholder",
+        }
+    )
+
+    def _resolve_effective_task(self, record: EpisodeRecord) -> str:
+        """Decide which task string drives the ``plan`` module for this episode.
+
+        Returns the user-supplied ``record.episode_task`` unless
+        ``derive_task_from_video`` says otherwise (see config docstring).
+        Falls back gracefully to the canonical task if video derivation
+        fails.
+        """
+        canonical = (record.episode_task or "").strip()
+        mode = (self.config.derive_task_from_video or "off").strip().lower()
+        if mode == "always":
+            derived = self._derive_task_from_video(record)
+            return derived or canonical
+        if mode == "if_short" and self._task_seems_bad(canonical):
+            derived = self._derive_task_from_video(record)
+            if derived:
+                return derived
+        return canonical
+
+    def _task_seems_bad(self, task: str) -> bool:
+        if not task:
+            return True
+        if len(task.split()) < int(self.config.derive_task_min_words):
+            return True
+        return task.lower() in self._PLACEHOLDER_TASKS
+
+    # ------------------------------------------------------------------
+    # VLM call helpers (factored out: every ``plan``-module prompt below follows
+    # the same "build messages → single VLM call → pull a named field"
+    # shape, only differing in field name + post-processing).
+    # ------------------------------------------------------------------
+
+    def _vlm_field(self, messages: list[dict[str, Any]], field: str) -> Any:
+        """Run a single VLM call and return ``result[field]`` or ``None``.
+
+        Centralizes the ``vlm.generate_json([m])[0]`` + ``isinstance(dict)``
+        dance every prompt-call site needs.
+        """
+        result = self.vlm.generate_json([messages])[0]
+        if isinstance(result, dict):
+            return result.get(field)
+        return None
+
+    @staticmethod
+    def _text_message(text: str) -> list[dict[str, Any]]:
+        """One-shot text-only user message wrapped for ``generate_json``."""
+        return [{"role": "user", "content": [{"type": "text", "text": text}]}]
+
+    def _video_message(
+        self,
+        record: EpisodeRecord,
+        prompt: str,
+        window: tuple[float, float] | None = None,
+    ) -> list[dict[str, Any]]:
+        """User message combining the (optionally windowed) video block with ``prompt``."""
+        content = [*self._episode_video_block(record, window=window), {"type": "text", "text": prompt}]
+        return [{"role": "user", "content": content}]
+
+    def _derive_task_from_video(self, record: EpisodeRecord) -> str | None:
+        """Ask the VLM "what is this video about" with no task hint at all."""
+        text = self._vlm_field(self._video_message(record, load_prompt("module_1_video_task")), "task")
+        return text.strip() if isinstance(text, str) and text.strip() else None
+
+    def _generate_task_rephrasings(self, base_task: str, *, n: int) -> list[str]:
+        """Generate ``n`` text-only paraphrases of ``base_task``."""
+        if n <= 0 or not base_task:
+            return []
+        prompt = load_prompt("module_1_task_rephrasings").format(base_task=base_task, n=n)
+        raw = self._vlm_field(self._text_message(prompt), "rephrasings")
+        if not isinstance(raw, list):
+            return []
+        out = [item.strip().strip('"').strip("'") for item in raw if isinstance(item, str)]
+        return [s for s in out if s][:n]
+
+    # ------------------------------------------------------------------
+    # Phase 1a + 1b: structured per-subtask action records
+    # ------------------------------------------------------------------
+
+    def _extract_action_record(
+        self,
+        record: EpisodeRecord,
+        span: dict[str, Any],
+        episode_task: str,
+    ) -> dict[str, Any] | None:
+        """Ask the VLM to extract a typed ``ActionRecord`` from a subtask span.
+
+        Sends ``frames_per_subtask`` frames uniformly sampled from
+        ``[span.start, span.end]`` plus the canonical subtask text. The
+        VLM is constrained to verb + grasp vocabularies from the config
+        — invalid values are silently dropped at this layer (the
+        validator catches structural problems pre-write).
+
+        Returns ``None`` when the call fails or the VLM returns something
+        unrecognizable; callers fall back to the free-form subtask text.
+        """
+        cfg = self.config.action_records
+        start_t = float(span.get("start", 0.0))
+        end_t = float(span.get("end", start_t))
+        duration = max(0.0, end_t - start_t)
+
+        # Uniform timestamps within the span; fall back to a single
+        # center frame for very short spans.
+        n = max(1, int(cfg.frames_per_subtask))
+        if n == 1 or duration <= 0.0:
+            timestamps = [0.5 * (start_t + end_t)]
+        else:
+            step = duration / (n - 1)
+            timestamps = [start_t + i * step for i in range(n)]
+        frames = self.frame_provider.frames_at(record, timestamps)
+        if not frames:
+            logger.debug(
+                "action_record: no frames at span %.2f-%.2f for ep %s; skipping",
+                start_t,
+                end_t,
+                record.episode_index,
+            )
+            return None
+
+        prompt = load_prompt("module_1_action_record").format(
+            episode_task=episode_task,
+            subtask_text=span.get("text", ""),
+            start_time=start_t,
+            end_time=end_t,
+            duration=duration,
+            n_frames=len(frames),
+            verb_vocabulary=", ".join(cfg.verb_vocabulary),
+            grasp_vocabulary=" | ".join(f'"{g}"' for g in cfg.grasp_vocabulary),
+        )
+        message = [
+            {
+                "role": "user",
+                "content": [*to_image_blocks(frames), {"type": "text", "text": prompt}],
+            }
+        ]
+        result = self.vlm.generate_json([message])[0]
+        if not isinstance(result, dict):
+            return None
+
+        # Light validation + normalisation. Verb is required; everything
+        # else may be null. Verb / grasp_type are clamped to the
+        # vocabularies (out-of-vocab → reject or null).
+        verb = (result.get("verb") or "").strip().lower()
+        if not verb or verb not in {v.lower() for v in cfg.verb_vocabulary}:
+            return None
+        obj = (result.get("object") or "").strip()
+        if not obj:
+            return None
+        grasp = result.get("grasp_type")
+        if isinstance(grasp, str):
+            grasp = grasp.strip().lower()
+            if grasp not in {g.lower() for g in cfg.grasp_vocabulary}:
+                grasp = None
+        else:
+            grasp = None
+        arm = result.get("arm")
+        if isinstance(arm, str):
+            arm = arm.strip().lower()
+            if arm not in {"left", "right", "both"}:
+                arm = None
+        else:
+            arm = None
+        destination = result.get("destination")
+        destination = destination.strip() if isinstance(destination, str) and destination.strip() else None
+        mistake = result.get("mistake")
+        mistake = mistake.strip() if isinstance(mistake, str) and mistake.strip() else None
+
+        return {
+            "verb": verb,
+            "object": obj,
+            "arm": arm,
+            "grasp_type": grasp,
+            "destination": destination,
+            "mistake": mistake,
+        }
+
+    # ------------------------------------------------------------------
+    # Structured 5-axis task augmentation (EgoMimic-style taxonomy)
+    # ------------------------------------------------------------------
+
+    def _generate_task_aug_by_axes(self, base_task: str, axes_cfg: Any) -> list[str]:
+        """One VLM call → variants along the 5-axis taxonomy.
+
+        Variants from all axes are flattened into a single list (the
+        downstream pipeline doesn't need to know about the per-axis
+        bucketing — every variant becomes a ``task_aug`` row). Order
+        is preserved for reproducibility: synonym_paraphrase first,
+        then omit_arm, then omit_orientation, then omit_grasp_method,
+        then combined_omissions.
+        """
+        if not base_task:
+            return []
+        prompt = load_prompt("module_1_task_aug_axes").format(
+            base_task=base_task,
+            n_synonym=axes_cfg.synonym_paraphrase,
+            n_omit_arm=axes_cfg.omit_arm,
+            n_omit_orientation=axes_cfg.omit_orientation,
+            n_omit_grasp_method=axes_cfg.omit_grasp_method,
+            n_combined=axes_cfg.combined_omissions,
+        )
+        result = self.vlm.generate_json([self._text_message(prompt)])[0]
+        if not isinstance(result, dict):
+            return []
+        ordered_axes = (
+            "synonym_paraphrase",
+            "omit_arm",
+            "omit_orientation",
+            "omit_grasp_method",
+            "combined_omissions",
+        )
+        flat: list[str] = []
+        seen: set[str] = set()
+        for axis in ordered_axes:
+            entries = result.get(axis)
+            if not isinstance(entries, list):
+                continue
+            for item in entries:
+                if not isinstance(item, str):
+                    continue
+                key = item.strip().strip('"').strip("'")
+                if not key or key in seen:
+                    continue
+                seen.add(key)
+                flat.append(key)
+        return flat
+
+    def _episode_video_block(
+        self, record: EpisodeRecord, window: tuple[float, float] | None = None
+    ) -> list[dict[str, Any]]:
+        """Video block for the segmentation / describe prompts.
+
+        Always returns a block that actually carries the video. When
+        ``use_video_url`` is set we try the server-side ``video_url``
+        path first, but if clip extraction fails we FALL BACK to
+        decoding + embedding frames rather than returning an empty
+        block — an empty block would leave the VLM with no visual
+        grounding at all and it would hallucinate subtasks purely from
+        the task text.
+
+        When ``window=(w0, w1)`` is given (windowed subtask generation,
+        ``subtask_window_seconds > 0``), embed frames sampled at the FIXED
+        ``frames_per_second`` rate within ``[w0, w1]`` — constant temporal
+        density regardless of episode length, so long episodes are split
+        into windows rather than subsampled to a sparse 32-frame whole-
+        episode view. The ``video_url`` path is skipped for windows (it is
+        a whole-episode clip). ``max_video_frames`` still caps each window
+        as a context-budget safety net.
+        """
+        if not record.frame_timestamps:
+            return []
+        if window is not None:
+            w0, w1 = float(window[0]), float(window[1])
+            dur = max(0.0, w1 - w0)
+            n = max(1, int(round(dur * self.config.frames_per_second)) + 1)
+            n = min(n, self.config.max_video_frames)
+            if n <= 1 or dur <= 0.0:
+                timestamps = [0.5 * (w0 + w1)]
+            else:
+                step = dur / (n - 1)
+                timestamps = [w0 + i * step for i in range(n)]
+            return to_video_block(self.frame_provider.frames_at(record, timestamps))
+        if self.config.use_video_url and isinstance(self.frame_provider, VideoFrameProvider):
+            cache_dir = Path(self.frame_provider.root) / ".annotate_staging" / ".video_clips"
+            clip = self.frame_provider.episode_clip_path(record, cache_dir)
+            if clip is not None:
+                return to_video_url_block(f"file://{clip}", fps=self.config.use_video_url_fps)
+            logger.warning(
+                "episode %d: video_url clip extraction failed — falling back to "
+                "embedded frames so the VLM still sees the demonstration",
+                record.episode_index,
+            )
+        episode_duration = record.frame_timestamps[-1] - record.frame_timestamps[0]
+        target_count = max(1, int(round(episode_duration * self.config.frames_per_second)))
+        target_count = min(target_count, self.config.max_video_frames)
+        video_frames = self.frame_provider.video_for_episode(record, target_count)
+        return to_video_block(video_frames)
+
+    def run_plan_updates(
+        self,
+        record: EpisodeRecord,
+        staging: EpisodeStaging,
+        interjection_times: Sequence[float],
+        interjection_texts: Sequence[str] | None = None,
+    ) -> None:
+        """Append additional ``plan`` rows at every interjection timestamp.
+
+        Plans refresh ONLY on user interjections — subtask generation
+        runs ~1 Hz at inference, but plan re-emission is event-driven.
+        Now also forwards the interjection's own text into the prompt so
+        the refreshed plan can actually reflect the user's correction
+        (the previous version told the model "an interjection happened"
+        without telling it what the user said).
+        """
+        if not self.config.emit_plan:
+            return
+        existing = staging.read("plan")
+        # Pass the episode's last frame timestamp so the final subtask
+        # span is closed (otherwise its ``end`` equals its ``start``,
+        # zero duration, and the "current subtask at refresh_t" lookup
+        # in ``_generate_plan`` misses any refresh that lands inside it).
+        episode_end_t = float(record.frame_timestamps[-1]) if record.frame_timestamps else None
+        spans = reconstruct_subtask_spans(existing, episode_end_t=episode_end_t)
+        already_planned: set[float] = {float(r["timestamp"]) for r in existing if r.get("style") == "plan"}
+        new_rows = list(existing)
+
+        texts: list[str | None] = (
+            [None] * len(interjection_times)
+            if interjection_texts is None
+            else [str(t) if t else None for t in interjection_texts]
+        )
+        for raw_t, inter_text in zip(interjection_times, texts, strict=True):
+            t = snap_to_frame(raw_t, record.frame_timestamps)
+            if t in already_planned:
+                continue
+            already_planned.add(t)
+            plan_text = self._generate_plan(record, spans, refresh_t=t, interjection=inter_text)
+            if plan_text is not None:
+                new_rows.append(
+                    {
+                        "role": "assistant",
+                        "content": plan_text,
+                        "style": "plan",
+                        "timestamp": t,
+                        "tool_calls": None,
+                    }
+                )
+        staging.write("plan", new_rows)
+
+    def _generate_subtasks(self, record: EpisodeRecord, *, task: str | None = None) -> list[dict[str, Any]]:
+        """Generate subtask spans, optionally via a multi-call quality chain.
+
+        Single call (default): watch video → emit subtask JSON.
+
+        Multi-call (opt-in, higher quality, more VLM calls):
+          1. ``subtask_describe_first`` — a grounding pass that narrates
+             ONLY what is visible (no JSON commitment to subtasks yet);
+             its description is injected into the segmentation prompt so
+             the model segments its own grounded observations instead of
+             pattern-matching the task text.
+          2. segmentation — emit subtask JSON (as before).
+        """
+        if record.row_count == 0 or not record.frame_timestamps:
+            return []
+        episode_duration = record.frame_timestamps[-1] - record.frame_timestamps[0]
+        effective_task = task if task is not None else record.episode_task
+
+        # ---- Windowed path (constant temporal density) ---------------
+        # When ``subtask_window_seconds > 0`` and the episode is longer
+        # than one window, process the episode in fixed-length windows so
+        # the VLM always sees ``frames_per_second`` density (instead of a
+        # sparse 32-frame whole-episode view). Each window runs the full
+        # describe -> segment chain on its own frames; results are merged +
+        # stitched into a contiguous whole-episode cover.
+        window_s = float(getattr(self.config, "subtask_window_seconds", 0.0) or 0.0)
+        if window_s > 0.0 and episode_duration > window_s:
+            return self._generate_subtasks_windowed(record, effective_task, window_s)
+
+        # ---- Pass 1 (optional): grounding description ----------------
+        observation_block = ""
+        if getattr(self.config, "subtask_describe_first", False):
+            description = self._describe_episode(record, effective_task)
+            if description:
+                observation_block = (
+                    "You watched this video and described, chronologically, "
+                    "ONLY what the robot actually does:\n"
+                    f'"""{description}"""\n\n'
+                    "Segment THAT grounded description (cross-checked against "
+                    "the video) into atomic subtasks. Do not introduce any "
+                    "action that is not in your description above.\n\n"
+                )
+
+        # ---- Pass 2: segmentation ------------------------------------
+        prompt = load_prompt("module_1_subtasks").format(
+            episode_task=effective_task,
+            min_subtask_seconds=self.config.min_subtask_seconds,
+            max_steps=self.config.plan_max_steps,
+            episode_duration=f"{episode_duration:.3f}",
+            observation_block=observation_block,
+        )
+        spans = self._vlm_field(self._video_message(record, prompt), "subtasks")
+        cleaned = self._clean_spans(spans, record)
+        if not cleaned:
+            return []
+
+        # ---- Full-episode coverage stitch ----------------------------
+        # The VLM can leave the first subtask starting after t0 or leave
+        # gaps between spans, so the subtask timeline no longer tiles the
+        # whole episode and frames fall through with no active subtask.
+        # Always stitch the surviving spans into a contiguous cover of
+        # [t0, t_last] — there is no scenario where a sparse, gap-ridden
+        # subtask timeline is desirable for conditioning.
+        cleaned = self._stitch_full_coverage(cleaned, record)
+
+        return cleaned
+
+    def _generate_subtasks_windowed(
+        self, record: EpisodeRecord, task: str, window_s: float
+    ) -> list[dict[str, Any]]:
+        """Subtask generation in fixed-length windows at constant fps.
+
+        Splits ``[t0, t_last]`` into consecutive windows of ``window_s``
+        seconds, runs the describe -> segment chain on each window's own
+        frames (sampled at ``frames_per_second``), offsets
+        each window's spans back to absolute episode time, then merges +
+        stitches into a contiguous whole-episode cover.
+        """
+        t0 = float(record.frame_timestamps[0])
+        t_last = float(record.frame_timestamps[-1])
+        all_spans: list[dict[str, Any]] = []
+        w0 = t0
+        n_windows = 0
+        while w0 < t_last - 1e-6:
+            w1 = min(w0 + window_s, t_last)
+            all_spans.extend(self._subtasks_for_window(record, task, w0, w1))
+            n_windows += 1
+            w0 = w1
+        logger.info(
+            "episode %d: windowed subtask gen over %d window(s) of %.1fs -> %d raw spans",
+            record.episode_index,
+            n_windows,
+            window_s,
+            len(all_spans),
+        )
+        # Merge across windows: clamp to the absolute episode, sort, and
+        # frame-snap to distinct starts (handles any boundary collisions).
+        cleaned = self._clean_spans(all_spans, record)
+        if not cleaned:
+            return []
+        return self._stitch_full_coverage(cleaned, record)
+
+    def _subtasks_for_window(
+        self, record: EpisodeRecord, task: str, w0: float, w1: float
+    ) -> list[dict[str, Any]]:
+        """Run describe -> segment on one ``[w0, w1]`` window.
+
+        The model works in window-RELATIVE time ``[0, L]`` (it perceives
+        the window as a clip starting at 0); spans are offset back to
+        absolute ``[w0, w1]`` before returning.
+        """
+        window = (w0, w1)
+        win_len = max(0.0, w1 - w0)
+
+        observation_block = ""
+        if getattr(self.config, "subtask_describe_first", False):
+            description = self._describe_episode(record, task, window=window)
+            if description:
+                observation_block = (
+                    "You watched this video clip and described, chronologically, "
+                    "ONLY what the robot actually does:\n"
+                    f'"""{description}"""\n\n'
+                    "Segment THAT grounded description (cross-checked against "
+                    "the clip) into atomic subtasks. Do not introduce any "
+                    "action that is not in your description above.\n\n"
+                )
+
+        prompt = load_prompt("module_1_subtasks").format(
+            episode_task=task,
+            min_subtask_seconds=self.config.min_subtask_seconds,
+            max_steps=self.config.plan_max_steps,
+            episode_duration=f"{win_len:.3f}",
+            observation_block=observation_block,
+        )
+        spans = self._vlm_field(self._video_message(record, prompt, window=window), "subtasks")
+        # Window-relative clamp; no frame-snap dedupe yet (done on the
+        # merged absolute set).
+        cleaned = self._clean_spans(spans, record, bounds=(0.0, win_len), dedupe=False)
+        if not cleaned:
+            return []
+
+        # Offset window-relative spans back to absolute episode time.
+        for s in cleaned:
+            s["start"] = w0 + float(s["start"])
+            s["end"] = w0 + float(s["end"])
+        return cleaned
+
+    def _stitch_full_coverage(
+        self, spans: list[dict[str, Any]], record: EpisodeRecord
+    ) -> list[dict[str, Any]]:
+        """Make subtask spans tile the full episode with no gaps.
+
+        * The first subtask starts at the episode's first frame ``t0``
+          (any idle / approach before the first labelled action is folded
+          into it), so every early frame has an active subtask.
+        * Each subtask's ``end`` is snapped to the next subtask's
+          ``start`` (gaps between spans are closed), and the final
+          subtask's ``end`` extends to the last frame ``t_last``.
+
+        Starts are otherwise left as the (already frame-snapped, distinct)
+        values the VLM produced — only the FIRST start is pulled
+        back to ``t0``, which can't collide with a later span because it
+        was already the earliest. Purely deterministic; runs after the
+        VLM passes.
+        """
+        if not spans or not record.frame_timestamps:
+            return spans
+        t0 = float(record.frame_timestamps[0])
+        t_last = float(record.frame_timestamps[-1])
+        spans = sorted(spans, key=lambda s: float(s["start"]))
+        spans[0]["start"] = t0
+        for i in range(len(spans) - 1):
+            spans[i]["end"] = float(spans[i + 1]["start"])
+        spans[-1]["end"] = t_last
+        for s in spans:
+            if float(s["end"]) < float(s["start"]):
+                s["end"] = float(s["start"])
+        return spans
+
+    def _clean_spans(
+        self,
+        spans: Any,
+        record: EpisodeRecord,
+        bounds: tuple[float, float] | None = None,
+        dedupe: bool = True,
+    ) -> list[dict[str, Any]]:
+        """Clamp / sort / (optionally) dedupe raw VLM subtask spans into valid rows.
+
+        ``bounds`` overrides the clamp range — pass the window's
+        ``(w_lo, w_hi)`` when cleaning window-relative spans, or leave
+        ``None`` to clamp to the whole episode ``[t0, t_last]``.
+        ``dedupe`` runs the frame-snap distinct-start step; skip it for
+        window-relative spans (frame snapping is done once on the merged,
+        absolute-time set).
+        """
+        if not spans:
+            return []
+        if bounds is not None:
+            lo, hi = float(bounds[0]), float(bounds[1])
+        else:
+            lo = record.frame_timestamps[0]
+            hi = record.frame_timestamps[-1]
+        cleaned: list[dict[str, Any]] = []
+        for span in spans:
+            try:
+                start = float(span["start"])
+                end = float(span["end"])
+                text = str(span["text"]).strip()
+            except (KeyError, ValueError, TypeError):
+                continue
+            start = max(lo, min(start, hi))
+            end = max(lo, min(end, hi))
+            if end < start:
+                start, end = end, start
+            if not text:
+                continue
+            cleaned.append({"text": text, "start": start, "end": end})
+        cleaned.sort(key=lambda s: s["start"])
+        if dedupe:
+            return self._dedupe_starts_to_distinct_frames(cleaned, record)
+        return cleaned
+
+    def _describe_episode(
+        self, record: EpisodeRecord, task: str, window: tuple[float, float] | None = None
+    ) -> str:
+        """Grounding pass: free-form chronological description of the (windowed) video."""
+        prompt = load_prompt("module_1_subtask_describe").format(episode_task=task)
+        text = self._vlm_field(self._video_message(record, prompt, window=window), "description")
+        return text.strip() if isinstance(text, str) and text.strip() else ""
+
+    @staticmethod
+    def _dedupe_starts_to_distinct_frames(
+        spans: list[dict[str, Any]], record: EpisodeRecord
+    ) -> list[dict[str, Any]]:
+        """Bump same-frame subtask starts onto distinct frames.
+
+        Two consecutive VLM spans whose ``start`` rounds to the same
+        source frame (after :func:`snap_to_frame`) would otherwise emit
+        two ``style=subtask`` rows at the identical persistent
+        timestamp. The training-time renderer's ``active_at(t,
+        style=subtask)`` resolver can't disambiguate that and raises
+        ``Ambiguous resolver for style='subtask'``.
+
+        Walk the (sorted-by-start) spans, snap each to its frame, and
+        if the snapped frame is already taken push the span onto the
+        next unused frame so both subtasks survive on distinct
+        timestamps. If the episode ends before a free frame is found,
+        the trailing span is dropped with a warning — better than
+        poisoning the render.
+        """
+        if not spans:
+            return spans
+        frames = record.frame_timestamps
+        if not frames:
+            return spans
+        used: set[float] = set()
+        out: list[dict[str, Any]] = []
+        for span in spans:
+            ts = snap_to_frame(span["start"], frames)
+            if ts in used:
+                next_ts = next((f for f in frames if f > ts and f not in used), None)
+                if next_ts is None:
+                    logger.warning(
+                        "episode %d: subtask %r snapped to occupied frame "
+                        "%.3f and no free later frame exists — dropping",
+                        record.episode_index,
+                        span.get("text"),
+                        ts,
+                    )
+                    continue
+                ts = next_ts
+            used.add(ts)
+            new_span = {**span, "start": ts}
+            if float(new_span.get("end", ts)) < ts:
+                new_span["end"] = ts
+            out.append(new_span)
+        return out
+
+    def _generate_plan(
+        self,
+        record: EpisodeRecord,  # noqa: ARG002  (kept for signature stability)
+        subtask_spans: Sequence[dict[str, Any]],
+        *,
+        refresh_t: float | None = None,
+        interjection: str | None = None,  # noqa: ARG002
+        task: str | None = None,  # noqa: ARG002
+    ) -> str | None:
+        """Deterministic plan = numbered list of *still-todo* subtasks.
+
+        Previously this called the VLM with a prompt that asked it to
+        compress the subtasks into a "compact hierarchical plan". That
+        produced longer-than-necessary plans, cost an extra VLM round-trip
+        per episode (plus one per interjection on refresh), and could
+        diverge from the actual subtask sequence the model is going to
+        execute. Replacing it with a plain summarisation keeps the plan
+        tightly aligned with the upcoming subtasks and removes the VLM
+        call entirely.
+
+        Layout — short imperative fragments prefixed by "N. ":
+
+            1. <subtask 1>
+            2. <subtask 2>
+            ...
+
+        On a refresh at ``refresh_t`` (called from ``run_plan_updates``
+        on interjection events, and from ``run_episode`` at every subtask
+        boundary), only subtasks whose start is at or after ``refresh_t``
+        are included — the plan shrinks as work progresses, so it always
+        describes what's left.
+        """
+        if not subtask_spans:
+            return None
+        remaining = [
+            s for s in subtask_spans if refresh_t is None or float(s.get("start", 0.0)) >= float(refresh_t)
+        ]
+        if not remaining:
+            # Past the last subtask boundary on a late refresh — nothing
+            # left to plan; emit None so the caller skips the row.
+            return None
+        return "\n".join(f"{i}. {span.get('text', '').strip()}" for i, span in enumerate(remaining, start=1))
+
+    def _generate_memory(
+        self,
+        record: EpisodeRecord,
+        prior_memory: str,
+        completed: str,
+        remaining: Sequence[str],
+        *,
+        task: str | None = None,
+    ) -> str:
+        prompt = load_prompt("module_1_memory").format(
+            episode_task=(task if task is not None else record.episode_task),
+            prior_memory=prior_memory or "(none)",
+            completed_subtask=completed,
+            remaining_subtasks=", ".join(remaining) if remaining else "(none)",
+        )
+        memory = self._vlm_field(self._text_message(prompt), "memory")
+        return memory.strip() if isinstance(memory, str) else ""

src/lerobot/annotations/steerable_pipeline/prompts/__init__.py

@@ -0,0 +1,33 @@
+#!/usr/bin/env python
+
+# Copyright 2026 The HuggingFace Inc. team. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""Prompt templates loaded as plain text.
+
+One file per use site. Templates use ``str.format(**vars)`` substitution; we
+intentionally avoid jinja2 here so the templates remain inspectable in
+plain editors and roundtrip cleanly through ``ruff format``.
+"""
+
+from __future__ import annotations
+
+from pathlib import Path
+
+_DIR = Path(__file__).parent
+
+
+def load(name: str) -> str:
+    """Read prompt template ``name.txt`` from the ``prompts/`` directory."""
+    path = _DIR / f"{name}.txt"
+    return path.read_text(encoding="utf-8")

src/lerobot/annotations/steerable_pipeline/prompts/module_1_action_record.txt

@@ -0,0 +1,64 @@
+You are extracting a structured action record from a subtask span of a
+teleoperated robot demonstration. This is Phase 1a of a two-step
+process: you extract a typed record; a deterministic template then
+renders it back to canonical subtask text. Your job is the PERCEPTION
+step — not the language step.
+
+The user originally asked: "{episode_task}"
+The subtask span is:        "{subtask_text}"
+Span time window:           [{start_time:.2f}s, {end_time:.2f}s]
+                            ({duration:.2f}s of robot activity)
+
+You are shown {n_frames} frames sampled uniformly from the subtask
+window. Fill in a structured record describing the action that takes
+place between the first and last frame.
+
+Hard rules:
+- Use ONLY information visible in the frames. Do not infer details from
+  outside the span. Do not extrapolate from the original task wording.
+- Use canonical object names from the original task VERBATIM. Never
+  introduce synonyms: if the task says "cube", the record says "cube",
+  never "block" / "object" / "item".
+- For non-applicable fields, use ``null`` (not "n/a", not "none", not
+  an empty string).
+- For ``verb`` and ``grasp_type``, pick EXACTLY one value from the
+  vocabulary below. Never invent a new one.
+
+Field schema:
+
+  verb (required) — the imperative verb of the action. Vocabulary:
+    {verb_vocabulary}
+
+  object (required) — the manipulated object. Use the canonical noun
+    from the original task above.
+
+  arm — which arm performs the action. One of:
+    "left" | "right" | "both" | null
+    Use ``null`` when the source robot is single-arm or when the arm
+    is genuinely not visible in the frames.
+
+  grasp_type — which grip the gripper uses on contact. One of:
+    {grasp_vocabulary} | null
+    Use ``null`` when there is no contact in this span (e.g. a pure
+    ``move`` / ``reach`` subtask) or the grip is genuinely unclear.
+
+  destination — the target location for actions like ``place``,
+    ``move``, ``insert``, ``pour``. Use canonical names from the
+    original task. Use ``null`` for in-place actions (``press``,
+    ``turn``, ``grasp``, ``release``).
+
+  mistake — a brief one-clause description of any visible failure or
+    recovery during the span (e.g. "dropped the cube and re-grasped",
+    "missed the target on first attempt"). Use ``null`` when the span
+    completes cleanly with no visible recovery.
+
+Output strictly valid JSON of shape:
+
+  {{
+    "verb": "<one of vocabulary>",
+    "object": "<canonical noun>",
+    "arm": "left" | "right" | "both" | null,
+    "grasp_type": "<one of vocabulary>" | null,
+    "destination": "<canonical noun>" | null,
+    "mistake": "<short description>" | null
+  }}

src/lerobot/annotations/steerable_pipeline/prompts/module_1_memory.txt

@@ -0,0 +1,36 @@
+You are updating the robot's compressed semantic memory at the boundary of
+a completed subtask.
+
+Reference (verbatim from MEM, Torne 2026):
+"Remove or compress information in the language memory whenever
+appropriate. Keep ONLY the minimal set of relevant information for future
+task execution. Specific object attributes (colors, precise quantities of
+each item) get discarded when their details won't affect subsequent
+actions. Functional outcomes (where items went, how many) are preserved."
+
+Episode task: "{episode_task}"
+Previous memory: {prior_memory}
+Just-completed subtask: "{completed_subtask}"
+Remaining subtasks (for relevance judgement only): {remaining_subtasks}
+
+Write the memory as a short FIRST-PERSON, PAST-TENSE narrative of what the
+robot has accomplished so far — the running story it would tell itself.
+
+Authoring rules:
+- First person, past tense. Every sentence starts with "I": "I picked
+  up...", "I opened...", "I moved to...".
+- One or two short sentences. Extend the previous memory with the
+  just-completed subtask; do not rewrite it from scratch.
+- Keep WHAT happened (functional outcomes — where items went, how many),
+  drop HOW (grasp details, motions).
+- Compress completed steps and drop object attributes (colors, exact
+  counts) once they no longer affect the remaining subtasks.
+
+Example (MEM, Torne 2026):
+  Before: "I prepared the pot and got the potatoes, milk, and butter. I
+           moved to the drawer."
+  After:  "I prepared the pot and got the ingredients. I opened the
+           drawer with the masher."
+
+Output strictly valid JSON:
+  {{ "memory": "<one or two short first-person past-tense sentences>" }}

src/lerobot/annotations/steerable_pipeline/prompts/module_1_subtask_describe.txt

@@ -0,0 +1,27 @@
+You are watching a teleoperated robot demonstration from a single
+camera. The user asked the robot to: "{episode_task}"
+
+This is an OBSERVATION pass. Watch the entire clip and describe, in
+chronological order, ONLY what the robot physically does — the concrete
+motions, approaches, contacts, grasps, releases, and relocations you can
+actually SEE in the frames.
+
+Hard rules:
+- Describe only motion visible in the video. Do NOT use the task
+  instruction to guess steps that aren't shown. The instruction is the
+  goal; the video is ground truth.
+- Do NOT segment into named subtasks yet and do NOT output JSON beyond
+  the single field below. Just narrate what happens.
+- Give an approximate timestamp (in seconds) for each distinct event,
+  e.g. "0.0-1.4s: the base drives forward toward the stove".
+- Do NOT invent objects, grasps, destinations, or steps. If the robot
+  only does one thing (e.g. it just navigates and the clip ends), say
+  exactly that and nothing more.
+- Be concrete and literal. "the gripper closes on the mug" — not "the
+  robot prepares to make coffee".
+
+Output strictly valid JSON:
+
+  {{
+    "description": "<chronological, timestamped description of ONLY what is visible>"
+  }}

src/lerobot/annotations/steerable_pipeline/prompts/module_1_subtasks.txt

@@ -0,0 +1,112 @@
+You are labeling a teleoperated robot demonstration.
+
+The user originally asked: "{episode_task}"
+
+You are shown the entire demonstration as a single video. Watch the
+whole clip, then segment it into a list of consecutive atomic subtasks
+the robot performs.
+
+{observation_block}GROUNDING — read this first, it overrides everything below:
+- Label ONLY what the robot actually does in the video. Every subtask
+  you emit must correspond to motion you can SEE in specific frames.
+- Do NOT invent, anticipate, or pad. If the robot only does one thing
+  (e.g. it just navigates to a location and the clip ends), emit
+  EXACTLY ONE subtask. Many demonstrations are a single atomic skill.
+- ``max_steps`` below is a hard CEILING, not a target. Emitting fewer
+  subtasks than the ceiling is not just allowed, it is expected for
+  short / atomic demonstrations. One correct subtask is far better
+  than several invented ones.
+- If the video does not clearly show the action implied by the task,
+  describe what you actually see — do NOT fabricate the task's steps
+  from the instruction text. The instruction tells you the goal; the
+  VIDEO is the ground truth for what happened.
+
+Authoring rules — Hi Robot atom granularity, pi0.7-style short prompts:
+
+- Each subtask = one COMPOSITE atomic skill the low-level policy can
+  execute end-to-end. A "skill" bundles its own approach motion with
+  its terminal action — do NOT split the approach off as its own
+  subtask. The whole-arm policy already learns to reach as part of
+  every manipulation primitive.
+- Write each subtask as an IMPERATIVE COMMAND, starting with one of
+  these verbs (extend only when none fits):
+    pick up <obj>           — approach + grasp + lift in one subtask
+    put <obj> on/in <loc>   — transport + release in one subtask
+    place <obj> on/in <loc> — synonym of "put"; pick one and stay consistent
+    push <obj>              — contact + linear shove
+    pull <obj>              — contact + linear retract
+    turn <knob/dial/handle> — rotary actuation
+    press <button>          — single-press contact
+    open <drawer/door/lid>  — full open motion
+    close <drawer/door/lid> — full close motion
+    pour <src> into <dst>   — tilt + flow
+    insert <obj> into <slot>— alignment + push-fit
+    go to <loc>             — ONLY when no grasp / actuation follows
+                             (e.g. a pure relocation between phases).
+                             If the next subtask grasps something at
+                             that location, drop "go to ..." and just
+                             write "pick up ..." instead.
+- Forbidden ultra-fine splits — the VLM is NOT allowed to emit these
+  as standalone subtasks; fold them into the parent composite:
+    "move to X"   → fold into "pick up X" (or whatever follows)
+    "reach for X" → fold into "pick up X"
+    "grasp X"     → fold into "pick up X"
+    "lift X"      → fold into "pick up X" (or "put X on Y" if it's
+                    the transport phase of a place)
+    "release X"   → fold into "put X on Y" (or "place X in Y")
+- Keep it SHORT — a verb phrase, not a sentence. Drop articles
+  ("the", "a") and adverbs ("carefully", "slowly"). Add a "how"
+  detail (which hand, which grasp point) ONLY when it is needed to
+  disambiguate. Every subtask must begin with one of the verbs
+  above (no leading nouns, no "then", no "first").
+- NEVER use third person. Never write "the robot", "the arm", "the
+  gripper moves", "it picks up" — the robot is implied. Command it,
+  do not describe it.
+- Use the exact object nouns from the task above. If the task says
+  "cube", every subtask says "cube" — never switch to "block". If it
+  says "box", never switch to "bin"/"container". Keep vocabulary
+  consistent across the whole episode.
+- Good: "pick up blue cube", "put blue cube in box", "open drawer",
+  "turn red knob", "press start button", "go to sink".
+- Bad: "move to blue cube" (approach as its own subtask — forbidden,
+  must be folded into "pick up blue cube"); "the robot arm moves
+  towards the blue cube" (third person, too long); "carefully pick
+  up the cube" (adverb, article); "release the yellow block"
+  ("block" when the task said "cube", and "release" must be folded
+  into a "put"/"place" subtask).
+- Subtasks are non-overlapping and cover the full episode in order.
+  Choose the cut points yourself based on what you see in the video
+  (gripper open/close events, contact, regrasps, transitions).
+- Each subtask spans at least {min_subtask_seconds} seconds. If a
+  candidate span would be shorter, merge it into its neighbour
+  rather than emitting it.
+- Do not exceed {max_steps} subtasks total. Fewer, larger composites
+  are preferred over many micro-steps.
+- Every subtask's [start_time, end_time] must lie within
+  [0.0, {episode_duration}] seconds.
+
+SPECIAL CASES — verb disambiguation (each rule is narrowly visual and
+fires ONLY on the spatial situation it names; it must not change how you
+label any other situation):
+- STACK vs PUT: if an object is placed ON TOP OF another specific object
+  (not on a flat table / shelf / counter), use "stack ... on ...", not
+  "put". "stack blue book on green book", NOT "put blue book on table".
+- INSERT vs PUT: if an object goes INTO a fitted slot / hole / socket /
+  receptacle (push-fit), use "insert ... into ...", not "put".
+- RETRIEVE/PICK-UP vs PUT (direction): watch the gripper. If it CLOSES
+  on the object and the object moves WITH the hand, it is "pick up" /
+  "retrieve" (object leaves its location). If the gripper OPENS and the
+  object stays where the hand left it, it is "put" / "place" (object
+  arrives at a location). Decide by which way the object moves, not by
+  where the hand ends up.
+- POUR vs PUT: only use "pour" when the source is tilted and contents
+  flow out; moving a full container without tilting is "put"/"place".
+
+Output strictly valid JSON of shape:
+
+  {{
+    "subtasks": [
+      {{"text": "<short imperative verb phrase>", "start": <float>, "end": <float>}},
+      ...
+    ]
+  }}

src/lerobot/annotations/steerable_pipeline/prompts/module_1_task_aug_axes.txt

@@ -0,0 +1,67 @@
+You are generating structured augmentations of a robot task instruction
+for training a language-conditioned policy. Unlike free-form rephrasing,
+your variants follow a NAMED 5-axis taxonomy — each axis omits or varies
+a specific element of the task while preserving its meaning.
+
+Original task: "{base_task}"
+
+Produce variants along five named axes. Each axis has a target count.
+The whole batch should expose the policy to maximum linguistic diversity
+WITHOUT changing what the robot is supposed to do.
+
+Axes and target counts:
+
+  synonym_paraphrase ({n_synonym}):
+    Different wording / verbs / sentence structure. ALL information
+    from the original task is preserved — same object, same arm
+    specification if present, same orientation if present, same grasp
+    if present.
+
+  omit_arm ({n_omit_arm}):
+    Drop the left/right/both arm specification from the task. Skip
+    entirely (emit 0 entries) if the original task does NOT mention an
+    arm. Do not invent an arm specification just to omit it.
+
+  omit_orientation ({n_omit_orientation}):
+    Drop orientation cues (upright, sideways, facing the user,
+    long-edge-first, etc.). Skip entirely if no orientation cue is
+    present in the original task.
+
+  omit_grasp_method ({n_omit_grasp_method}):
+    Drop the grip / grasp method specification (pinch, wrap, hold by
+    the rim, etc.). Skip entirely if no grasp method is mentioned.
+
+  combined_omissions ({n_combined}):
+    Combine TWO of the above omissions simultaneously (e.g. drop both
+    arm and orientation). Skip entirely if fewer than two of (arm,
+    orientation, grasp_method) appear in the original task.
+
+Hard rules:
+- Each variant MUST preserve the core action, the target object, AND
+  the goal / destination. Do not change which object is involved, where
+  it goes, or the high-level action. "Navigate to the stove" may become
+  "go to the stove" or "head over to the stove" — it must NEVER become
+  "wander around the kitchen", "explore the room", or anything that
+  drops or generalises the stove destination. If you cannot vary the
+  wording without changing the goal, emit fewer variants.
+- Only the FIVE listed elements (wording, arm, orientation, grasp
+  method, or a combination) may be varied or omitted. The verb's
+  meaning, the object, and the destination are fixed.
+- Each variant is plain prose, no markdown, no quotes, no list numbers.
+- Each variant must be DISTINCT from every other variant in the entire
+  output, both within and across axes. Near-duplicates are not allowed.
+- If an axis cannot reach its target count because the original task
+  lacks the omittable element, emit fewer entries — do NOT pad the
+  axis with paraphrases that belong to a different axis.
+- Variants should not all start with verbs — vary sentence structure
+  (some imperative, some polite request, some question).
+
+Output strictly valid JSON of shape:
+
+  {{
+    "synonym_paraphrase": ["<v1>", "<v2>", ...],
+    "omit_arm": ["<v1>", "<v2>", ...],
+    "omit_orientation": ["<v1>", ...],
+    "omit_grasp_method": ["<v1>", ...],
+    "combined_omissions": ["<v1>", ...]
+  }}

src/lerobot/annotations/steerable_pipeline/prompts/module_1_task_rephrasings.txt

@@ -0,0 +1,32 @@
+You are generating training data for a Hi Robot-style policy. We need
+{n} alternative phrasings of the same robot task so the policy sees
+diverse user prompts during training instead of the same canonical
+string repeated every frame.
+
+Original task:
+"{base_task}"
+
+Generate exactly {n} alternative phrasings of the same task. Vary:
+
+- formality (casual / polite / curt)
+- verbosity (mostly short imperative; occasional polite request)
+- word choice (synonyms, different verbs)
+- sentence structure (imperative / question / suggestion)
+
+Hard rules:
+- Each phrasing MUST preserve the exact meaning of the original task.
+  Do not change which object is involved, the destination, or the
+  action. Do not add extra steps. Do not invent new objects.
+- Each phrasing must be a short phrase or sentence, plain prose, no
+  markdown, no quotes, no list numbers.
+- Phrasings must be distinct — no near-duplicates.
+- Output exactly {n} entries.
+
+Output strictly valid JSON:
+  {{
+    "rephrasings": [
+      "<phrasing 1>",
+      "<phrasing 2>",
+      ...
+    ]
+  }}

src/lerobot/annotations/steerable_pipeline/prompts/module_1_video_task.txt

@@ -0,0 +1,17 @@
+The video above shows a robot manipulation episode in full. Look at
+the entire video and describe in ONE concise sentence what the robot
+is doing.
+
+Rules:
+- One sentence, in natural English, like a user instruction.
+- Capture the goal of the demonstration, not low-level motions.
+  Example: "place the yellow cube into the red bin" — not "move the
+  end-effector down 5cm and close the gripper".
+- 4 to 15 words. Plain prose, no markdown, no bullets, no quotes.
+- Do not invent objects or actions that aren't visible.
+- Do not output anything other than the JSON object below.
+
+Output strictly valid JSON:
+  {{
+    "task": "<single concise sentence describing what the robot does in this video>"
+  }}

src/lerobot/annotations/steerable_pipeline/prompts/module_2_initial_speech.txt

@@ -0,0 +1,12 @@
+The user just asked the robot: "{episode_task}".
+
+Generate a short verbal acknowledgement the robot would speak back before
+beginning the task. Style: compact, confident, friendly.
+
+Examples (Hi Robot, Shi 2025): "Sure, I won't put cheese on it.",
+"OK, starting with the sponge.", "Got it.".
+
+Prefer very short replies: "Got it.", "On it.", "OK."
+
+Output strictly valid JSON:
+  {{ "text": "<the spoken acknowledgement>" }}

src/lerobot/annotations/steerable_pipeline/prompts/module_2_interjection.txt

@@ -0,0 +1,46 @@
+You are generating training data for a Hi Robot-style hierarchical
+robot policy. The robot in this demonstration has ALREADY executed
+every step shown in the video — we cannot retroactively change the
+action stream. To keep training data consistent with the video, the
+"interjection" must align with what the robot is *about to do next* in
+the demonstration, framed as a natural mid-task user request.
+
+The episode's overall task: "{episode_task}".
+
+The images above show roughly {window_seconds:.1f} seconds straddling a
+subtask boundary in the demonstration:
+
+- Subtask the robot just finished: "{prev_subtask}"
+- Subtask the robot is about to start: "{next_subtask}"
+- Time into episode: {timestamp:.2f}s
+
+Write ONE compact interjection the user would naturally say at this
+moment to prompt / confirm / encourage the robot to do "{next_subtask}".
+Keep it like a mid-task coaching cue, not a full instruction paragraph.
+Also write the robot's compact verbal acknowledgement.
+
+Hard rules:
+
+- The interjection MUST be consistent with the next subtask. The user
+  cannot ask for something different from what the robot then does in
+  the video. If you're tempted to say "actually skip X" or "do Y
+  instead", DO NOT — those would contradict the demonstration.
+- The interjection must reference an object, location, or action that
+  is plausible given the visible scene and the next subtask text.
+- One short phrase or sentence each. Conversational, not robotic.
+- Prefer direct cues: "{next_subtask}, please."; "Now {next_subtask}."
+- Keep robot speech very short: "OK.", "On it.", "Doing that."
+
+Style examples (vary the phrasing — don't reuse these verbatim):
+  - "Now go ahead and {next_subtask}."
+  - "Great, can you {next_subtask} next?"
+  - "{next_subtask}, please."
+  - "Before you continue, please {next_subtask}."
+  - "Looking good — {next_subtask} now."
+  - "Okay, {next_subtask}."
+
+Output strictly valid JSON:
+  {{
+    "interjection": "<short cue from the user, asking for the next subtask>",
+    "speech":       "<short robot acknowledgement>"
+  }}

src/lerobot/annotations/steerable_pipeline/prompts/module_3_vqa.txt

@@ -0,0 +1,32 @@
+You are generating a frame-grounded visual question/answer pair for
+chain-of-thought training. Reference: ECoT (Zawalski 2024) and Steerable
+Policies — both train policies on grounded features such as bounding box
+pixel coordinates, keypoints, counts, attributes, and spatial relations.
+
+The frame shows a robot working on: "{episode_task}".
+
+Question types and the EXACT answer JSON shape required for each:
+
+  bbox       => {{"detections": [{{"label": "<obj>", "bbox_format": "xyxy",
+                                    "bbox": [x1, y1, x2, y2]}}, ...]}}
+                bbox is in pixel coordinates (x_min, y_min, x_max, y_max).
+                ECoT example: "a white cup [124, 25, 176, 113]".
+
+  keypoint   => {{"label": "<point>", "point_format": "xy",
+                  "point": [x, y]}}
+
+  count      => {{"label": "<obj>", "count": <int>,
+                  "note": "<optional short note>"}}
+
+  attribute  => {{"label": "<obj>", "attribute": "<color|shape|state|...>",
+                  "value": "<observed value>"}}
+
+  spatial    => {{"subject": "<obj>", "relation": "<left_of|right_of|on|in|"
+                  "above|below|near>", "object": "<obj>"}}
+
+Generate a question of type "{question_type}". Output strictly valid JSON:
+
+  {{
+    "question": "<short, frame-grounded question>",
+    "answer":   <object whose shape matches the schema above>
+  }}

src/lerobot/annotations/steerable_pipeline/reader.py

@@ -0,0 +1,274 @@
+#!/usr/bin/env python
+
+# Copyright 2026 The HuggingFace Inc. team. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""Datatrove-shaped reader.
+
+The reader walks ``data/chunk-*/file-*.parquet`` and yields one record per
+episode containing:
+
+- ``episode_index``: int
+- ``frame_timestamps``: tuple[float, ...]
+- ``frame_indices``: tuple[int, ...]
+- ``episode_task``: str (canonical task from ``meta/tasks.parquet``)
+- ``data_path``: pathlib.Path of the source parquet shard
+- ``frames_df``: pandas.DataFrame slice for the episode (only loaded on demand)
+
+This shape lets each module operate per-episode without loading all parquet
+rows into memory at once.
+"""
+
+from __future__ import annotations
+
+from collections.abc import Iterator, Sequence
+from dataclasses import dataclass, field
+from pathlib import Path
+from typing import Any
+
+import pyarrow.parquet as pq
+
+from lerobot.datasets.io_utils import load_tasks
+from lerobot.datasets.utils import DEFAULT_TASKS_PATH
+
+
+@dataclass
+class EpisodeRecord:
+    """Per-episode record yielded by the reader."""
+
+    episode_index: int
+    episode_task: str
+    frame_timestamps: tuple[float, ...]
+    frame_indices: tuple[int, ...]
+    data_path: Path
+    row_offset: int  # row offset within the parquet file where this episode starts
+    row_count: int  # number of rows for this episode
+
+    # Memoized parquet slice — populated on first ``frames_df()`` call so
+    # repeat queries from different modules don't re-read the whole shard.
+    _frames_df_cache: Any = field(default=None, init=False, repr=False, compare=False)
+
+    def frames_df(self):  # type: ignore[no-untyped-def]
+        """Lazy-load the pandas slice for this episode (memoized)."""
+        if self._frames_df_cache is None:
+            import pandas as pd  # noqa: PLC0415  - deferred for optional dataset extra
+
+            table = pq.read_table(self.data_path)
+            df: pd.DataFrame = table.to_pandas()
+            self._frames_df_cache = df.iloc[self.row_offset : self.row_offset + self.row_count].reset_index(
+                drop=True
+            )
+        return self._frames_df_cache
+
+
+def reconstruct_subtask_spans(
+    rows: Sequence[dict[str, Any]],
+    *,
+    episode_end_t: float | None = None,
+) -> list[dict[str, Any]]:
+    """Turn ``style="subtask"`` rows into ``{text, start, end}`` spans.
+
+    Each span's ``end`` is the next span's ``start``. The final span's
+    ``end`` defaults to its own ``start`` (zero-duration) — pass
+    ``episode_end_t`` to extend it to the episode's last frame instead,
+    which is what downstream consumers (memory, interjection boundary
+    selection) expect.
+
+    Used by the ``plan`` module (plan-update pass) and the
+    ``interjections`` module (interjection anchoring), which both need the
+    same span shape.
+    """
+    sorted_rows = sorted(
+        (r for r in rows if r.get("style") == "subtask"),
+        key=lambda r: float(r["timestamp"]),
+    )
+    spans: list[dict[str, Any]] = []
+    for r in sorted_rows:
+        t = float(r["timestamp"])
+        if spans:
+            spans[-1]["end"] = t
+        spans.append({"text": r.get("content") or "", "start": t, "end": t})
+    if spans and episode_end_t is not None and float(episode_end_t) > spans[-1]["start"]:
+        spans[-1]["end"] = float(episode_end_t)
+    return spans
+
+
+def snap_to_frame(t: float, frame_timestamps: Sequence[float]) -> float:
+    """Snap an arbitrary float to the nearest exact source frame timestamp.
+
+    Modules use this when emitting event-style rows so the row's
+    timestamp matches a real parquet frame: event rows must land on an
+    exact frame, otherwise the per-frame event lookup the writer does
+    would never match them.
+    """
+    if not frame_timestamps:
+        return float(t)
+    nearest = min(frame_timestamps, key=lambda f: abs(f - t))
+    return float(nearest)
+
+
+def _load_tasks_lookup(root: Path) -> dict[int, str]:
+    """Map ``task_index -> task`` from ``meta/tasks.parquet``.
+
+    Returns an empty dict when the file is absent — the task description is
+    derived later from the video if needed. Reuses the library-level
+    :func:`lerobot.datasets.io_utils.load_tasks`, which returns the tasks
+    frame indexed by task string with a ``task_index`` column.
+    """
+    if not (root / DEFAULT_TASKS_PATH).exists():
+        return {}
+    tasks = load_tasks(root)
+    return {int(idx): str(task) for task, idx in zip(tasks.index, tasks["task_index"], strict=True)}
+
+
+def iter_episodes(root: Path, *, only_episodes: tuple[int, ...] | None = None) -> Iterator[EpisodeRecord]:
+    """Yield :class:`EpisodeRecord` for every episode under ``root/data/``.
+
+    Episodes are yielded in ascending ``episode_index`` order. The reader does
+    not assume a specific chunk/file layout: it scans every ``*.parquet``
+    under ``data/`` and groups by ``episode_index``.
+    """
+    tasks = _load_tasks_lookup(root)
+    data_dir = root / "data"
+    parquet_files = sorted(data_dir.rglob("*.parquet"))
+
+    only_set = set(only_episodes) if only_episodes is not None else None
+
+    for path in parquet_files:
+        yield from _iter_one_path(path, tasks, only_set)
+
+
+def _iter_one_path(path: Path, tasks: dict[int, str], only_set: set[int] | None) -> Iterator[EpisodeRecord]:
+    table = pq.read_table(path)
+    names = table.column_names
+    if "episode_index" not in names:
+        return
+    episode_col = table.column("episode_index").to_pylist()
+    timestamp_col = (
+        table.column("timestamp").to_pylist() if "timestamp" in names else [0.0] * len(episode_col)
+    )
+    frame_col = (
+        table.column("frame_index").to_pylist() if "frame_index" in names else list(range(len(episode_col)))
+    )
+    task_col = table.column("task_index").to_pylist() if "task_index" in names else None
+
+    def _build(
+        ep: int,
+        start: int,
+        end: int,
+        task_idx: int | None,
+        ts_buf: list[float],
+        fi_buf: list[int],
+    ) -> EpisodeRecord | None:
+        if only_set is not None and ep not in only_set:
+            return None
+        task = tasks.get(task_idx, "") if task_idx is not None else ""
+        return EpisodeRecord(
+            episode_index=ep,
+            episode_task=task,
+            frame_timestamps=tuple(ts_buf),
+            frame_indices=tuple(fi_buf),
+            data_path=path,
+            row_offset=start,
+            row_count=end - start,
+        )
+
+    cur_ep: int | None = None
+    start_offset = 0
+    ts_buf: list[float] = []
+    fi_buf: list[int] = []
+    cur_task_idx: int | None = None
+
+    for i, ep in enumerate(episode_col):
+        if cur_ep is None:
+            cur_ep = ep
+            start_offset = i
+            ts_buf = [timestamp_col[i]]
+            fi_buf = [frame_col[i]]
+            cur_task_idx = task_col[i] if task_col is not None else None
+            continue
+        if ep != cur_ep:
+            rec = _build(cur_ep, start_offset, i, cur_task_idx, ts_buf, fi_buf)
+            if rec is not None:
+                yield rec
+            cur_ep = ep
+            start_offset = i
+            ts_buf = [timestamp_col[i]]
+            fi_buf = [frame_col[i]]
+            cur_task_idx = task_col[i] if task_col is not None else None
+        else:
+            ts_buf.append(timestamp_col[i])
+            fi_buf.append(frame_col[i])
+
+    if cur_ep is not None:
+        rec = _build(cur_ep, start_offset, len(episode_col), cur_task_idx, ts_buf, fi_buf)
+        if rec is not None:
+            yield rec
+
+
+def gather_data_paths(root: Path) -> list[Path]:
+    """Return every ``data/chunk-*/file-*.parquet`` path under ``root``."""
+    return sorted((root / "data").rglob("*.parquet"))
+
+
+def episode_offsets_per_path(path: Path) -> dict[int, tuple[int, int]]:
+    """Return ``{episode_index: (row_offset, row_count)}`` for one parquet."""
+    table = pq.read_table(path, columns=["episode_index"])
+    episode_col = table.column("episode_index").to_pylist()
+    out: dict[int, tuple[int, int]] = {}
+    cur_ep: int | None = None
+    start = 0
+    for i, ep in enumerate(episode_col):
+        if cur_ep is None:
+            cur_ep = ep
+            start = i
+            continue
+        if ep != cur_ep:
+            out[cur_ep] = (start, i - start)
+            cur_ep = ep
+            start = i
+    if cur_ep is not None:
+        out[cur_ep] = (start, len(episode_col) - start)
+    return out
+
+
+def keyframe_indices(record: EpisodeRecord, k: int) -> list[int]:
+    """Return ``k`` evenly spaced row indices into the episode (relative)."""
+    n = record.row_count
+    if k <= 0 or n == 0:
+        return []
+    if k >= n:
+        return list(range(n))
+    step = (n - 1) / (k - 1) if k > 1 else 0.0
+    return [int(round(i * step)) for i in range(k)] if k > 1 else [n // 2]
+
+
+def lookup_data_path(root: Path, episode_index: int) -> tuple[Path, int, int] | None:
+    """Find the parquet file containing ``episode_index`` and its slice bounds."""
+    for path in gather_data_paths(root):
+        offsets = episode_offsets_per_path(path)
+        if episode_index in offsets:
+            start, count = offsets[episode_index]
+            return path, start, count
+    return None
+
+
+def episode_frame_timestamps(root: Path, episode_index: int) -> tuple[Any, list[float]]:
+    """Return the parquet path and per-frame timestamps for ``episode_index``."""
+    found = lookup_data_path(root, episode_index)
+    if found is None:
+        raise ValueError(f"Episode {episode_index} not found under {root}/data/")
+    path, start, count = found
+    table = pq.read_table(path, columns=["timestamp"])
+    timestamps = table.column("timestamp").to_pylist()[start : start + count]
+    return path, [float(t) for t in timestamps]

src/lerobot/annotations/steerable_pipeline/staging.py

@@ -0,0 +1,104 @@
+#!/usr/bin/env python
+
+# Copyright 2026 The HuggingFace Inc. team. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""Per-episode staging.
+
+Each module writes its raw output as a JSONL file under
+``<staging_dir>/episode_{ep:06d}/<module>.jsonl``. The writer reads back this
+staging tree and partitions rows into the two language columns.
+
+JSONL is preferred over parquet here because the staging artifact is meant to
+be human-inspectable, easy to diff between prompt iterations, and trivially
+appended to. The final dataset format is parquet; staging is just an
+intermediate.
+"""
+
+from __future__ import annotations
+
+import json
+from collections.abc import Iterable, Iterator
+from dataclasses import dataclass
+from pathlib import Path
+from typing import Any
+
+ModuleName = str
+
+_MODULES: tuple[ModuleName, ...] = (
+    "plan",
+    "interjections",
+    "vqa",
+)
+
+
+@dataclass
+class EpisodeStaging:
+    """Filesystem layout for a single episode's staged module outputs."""
+
+    root: Path
+    episode_index: int
+
+    @property
+    def episode_dir(self) -> Path:
+        return self.root / f"episode_{self.episode_index:06d}"
+
+    def path_for(self, module: ModuleName) -> Path:
+        if module not in _MODULES:
+            raise ValueError(f"Unknown module {module!r}; expected one of {_MODULES}")
+        return self.episode_dir / f"{module}.jsonl"
+
+    def write(self, module: ModuleName, rows: Iterable[dict[str, Any]]) -> Path:
+        path = self.path_for(module)
+        path.parent.mkdir(parents=True, exist_ok=True)
+        # Atomic replace: a crash mid-write would otherwise leave a
+        # half-written JSONL file that ``read()`` would then fail to
+        # parse. Write to a sibling .tmp and rename so the target path
+        # only ever points at a complete file.
+        tmp_path = path.with_suffix(path.suffix + ".tmp")
+        with tmp_path.open("w", encoding="utf-8") as f:
+            for row in rows:
+                f.write(json.dumps(row, ensure_ascii=False, sort_keys=True))
+                f.write("\n")
+        tmp_path.replace(path)
+        return path
+
+    def read(self, module: ModuleName) -> list[dict[str, Any]]:
+        path = self.path_for(module)
+        if not path.exists():
+            return []
+        out: list[dict[str, Any]] = []
+        with path.open(encoding="utf-8") as f:
+            for line in f:
+                line = line.strip()
+                if line:
+                    out.append(json.loads(line))
+        return out
+
+    def read_all(self) -> dict[ModuleName, list[dict[str, Any]]]:
+        return {m: self.read(m) for m in _MODULES}
+
+    def has(self, module: ModuleName) -> bool:
+        return self.path_for(module).exists()
+
+
+def iter_staged_episodes(root: Path) -> Iterator[int]:
+    """Yield episode indices for which any staging artifact exists."""
+    if not root.exists():
+        return
+    for child in sorted(root.iterdir()):
+        if child.is_dir() and child.name.startswith("episode_"):
+            try:
+                yield int(child.name.removeprefix("episode_"))
+            except ValueError:
+                continue

src/lerobot/annotations/steerable_pipeline/validator.py

@@ -0,0 +1,326 @@
+#!/usr/bin/env python
+
+# Copyright 2026 The HuggingFace Inc. team. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""Pre-write validation against staged outputs.
+
+Runs after all three modules have written their per-episode artifacts but
+*before* the writer rewrites parquet shards. The validator never touches
+parquet; it only inspects the staging tree and the source frame timestamps
+exposed by :class:`EpisodeRecord`.
+
+Checks (per the plan's "Intermediate staging and validation" section):
+
+- exact timestamp alignment against source frame timestamps
+- no orphan speech / interjection pairs
+- plan / memory emission consistency (events have a paired persistent row)
+- VQA assistant ``content`` is valid JSON (one of bbox / keypoint / count /
+  attribute / spatial)
+- every row maps to its correct column under :func:`column_for_style`
+"""
+
+from __future__ import annotations
+
+import json
+import logging
+from collections.abc import Iterable, Sequence
+from dataclasses import dataclass, field
+from pathlib import Path
+from typing import Any
+
+from lerobot.datasets.language import (
+    LANGUAGE_EVENTS,
+    LANGUAGE_PERSISTENT,
+    column_for_style,
+    is_view_dependent_style,
+    validate_camera_field,
+)
+
+from .reader import EpisodeRecord
+from .staging import EpisodeStaging
+
+logger = logging.getLogger(__name__)
+
+
+@dataclass
+class ValidationReport:
+    """Outcome of one validation pass across all episodes."""
+
+    errors: list[str] = field(default_factory=list)
+    warnings: list[str] = field(default_factory=list)
+    episodes_checked: int = 0
+
+    @property
+    def ok(self) -> bool:
+        return not self.errors
+
+    def add_error(self, message: str) -> None:
+        self.errors.append(message)
+
+    def add_warning(self, message: str) -> None:
+        self.warnings.append(message)
+
+    def summary(self) -> str:
+        return f"checked={self.episodes_checked} errors={len(self.errors)} warnings={len(self.warnings)}"
+
+
+VQA_ANSWER_SHAPES: dict[str, set[str]] = {
+    "bbox": {"detections"},
+    "keypoint": {"label", "point_format", "point"},
+    "count": {"label", "count"},
+    "attribute": {"label", "attribute", "value"},
+    "spatial": {"subject", "relation", "object"},
+}
+
+
+def classify_vqa_answer(payload: Any) -> str | None:
+    """Best-effort classification of a VQA answer payload to a question type."""
+    if not isinstance(payload, dict):
+        return None
+    keys = set(payload.keys())
+    for kind, required in VQA_ANSWER_SHAPES.items():
+        if required.issubset(keys):
+            return kind
+    return None
+
+
+@dataclass
+class StagingValidator:
+    """Walks the staging tree and produces a :class:`ValidationReport`."""
+
+    timestamp_atol: float = 0.0  # exact-match by default
+    dataset_camera_keys: tuple[str, ...] | None = None
+    """Known ``observation.images.*`` keys on the dataset. When set, the
+    validator additionally enforces that every view-dependent row's
+    ``camera`` field references one of these keys. Pass ``None`` (default)
+    to skip that cross-check (e.g. in unit tests with no real dataset)."""
+
+    def validate(
+        self,
+        records: Sequence[EpisodeRecord],
+        staging_dir: Path,
+    ) -> ValidationReport:
+        report = ValidationReport()
+        for record in records:
+            self._validate_episode(record, staging_dir, report)
+            report.episodes_checked += 1
+        return report
+
+    def _validate_episode(
+        self,
+        record: EpisodeRecord,
+        staging_dir: Path,
+        report: ValidationReport,
+    ) -> None:
+        staging = EpisodeStaging(staging_dir, record.episode_index)
+        staged = staging.read_all()
+        all_rows: list[dict[str, Any]] = []
+        for module_name, rows in staged.items():
+            for row in rows:
+                row = {**row, "_module": module_name}
+                all_rows.append(row)
+
+        frame_ts = set(record.frame_timestamps)
+
+        events: list[dict[str, Any]] = []
+        persistent: list[dict[str, Any]] = []
+        for row in all_rows:
+            self._check_column_routing(row, report, record.episode_index)
+            self._check_camera_field(row, report, record.episode_index, self.dataset_camera_keys)
+            if column_for_style(row.get("style")) == LANGUAGE_PERSISTENT:
+                persistent.append(row)
+            else:
+                events.append(row)
+
+        for row in events:
+            self._check_event_timestamp_alignment(row, frame_ts, report, record.episode_index)
+
+        self._check_speech_interjection_pairs(events, report, record.episode_index)
+        self._check_plan_memory_consistency(persistent, events, report, record.episode_index)
+        self._check_vqa_json(events, report, record.episode_index)
+        self._check_vqa_uniqueness_per_frame_camera(events, report, record.episode_index)
+
+    def _check_camera_field(
+        self,
+        row: dict[str, Any],
+        report: ValidationReport,
+        episode_index: int,
+        dataset_camera_keys: Sequence[str] | None,
+    ) -> None:
+        """Enforce the camera invariant + that the key matches the dataset's cameras."""
+        style = row.get("style")
+        camera = row.get("camera")
+        try:
+            validate_camera_field(style, camera)
+        except ValueError as exc:
+            report.add_error(f"ep={episode_index} module={row.get('_module')}: {exc}")
+            return
+        if is_view_dependent_style(style) and dataset_camera_keys and camera not in dataset_camera_keys:
+            report.add_error(
+                f"ep={episode_index} module={row.get('_module')}: camera {camera!r} on style "
+                f"{style!r} is not one of the dataset's video keys {sorted(dataset_camera_keys)!r}"
+            )
+
+    def _check_vqa_uniqueness_per_frame_camera(
+        self,
+        events: Iterable[dict[str, Any]],
+        report: ValidationReport,
+        episode_index: int,
+    ) -> None:
+        """Ensure at most one (vqa, user) and one (vqa, assistant) per (t, camera)."""
+        counts: dict[tuple[float, str, str], int] = {}
+        for row in events:
+            if row.get("style") != "vqa":
+                continue
+            ts = row.get("timestamp")
+            camera = row.get("camera")
+            role = row.get("role")
+            if ts is None or camera is None or role is None:
+                continue  # other validators flag these
+            key = (float(ts), str(camera), str(role))
+            counts[key] = counts.get(key, 0) + 1
+        for (ts, camera, role), n in counts.items():
+            if n > 1:
+                report.add_error(
+                    f"ep={episode_index}: {n} duplicate vqa rows at t={ts} "
+                    f"camera={camera!r} role={role!r}; expected at most one per (t, camera, role)"
+                )
+
+    def _check_column_routing(
+        self,
+        row: dict[str, Any],
+        report: ValidationReport,
+        episode_index: int,
+    ) -> None:
+        style = row.get("style")
+        module = row.get("_module")
+        try:
+            target_col = column_for_style(style)
+        except ValueError:
+            report.add_error(f"ep={episode_index} module={module}: unknown style {style!r}")
+            return
+        if module == "plan" and target_col != LANGUAGE_PERSISTENT:
+            report.add_error(
+                f"ep={episode_index} module=plan emitted style {style!r} that routes to {target_col} (must be persistent)"
+            )
+        if module in {"interjections", "vqa"} and target_col != LANGUAGE_EVENTS:
+            report.add_error(
+                f"ep={episode_index} module={module} emitted style {style!r} that routes to {target_col} (must be events)"
+            )
+
+    def _check_event_timestamp_alignment(
+        self,
+        row: dict[str, Any],
+        frame_ts: set[float],
+        report: ValidationReport,
+        episode_index: int,
+    ) -> None:
+        ts = row.get("timestamp")
+        if ts is None:
+            report.add_error(f"ep={episode_index}: event row missing timestamp: {row!r}")
+            return
+        if self.timestamp_atol == 0.0:
+            if float(ts) not in frame_ts:
+                report.add_error(
+                    f"ep={episode_index}: event row timestamp {ts!r} does not match any source frame timestamp"
+                )
+        else:
+            if not any(abs(float(ts) - f) <= self.timestamp_atol for f in frame_ts):
+                report.add_error(
+                    f"ep={episode_index}: event row timestamp {ts!r} not within {self.timestamp_atol}s of any frame"
+                )
+
+    def _check_speech_interjection_pairs(
+        self,
+        events: Iterable[dict[str, Any]],
+        report: ValidationReport,
+        episode_index: int,
+    ) -> None:
+        speech_ts: dict[float, int] = {}
+        interjection_ts: dict[float, int] = {}
+        for row in events:
+            ts = row.get("timestamp")
+            if ts is None:
+                continue
+            ts_f = float(ts)
+            if row.get("style") is None and row.get("role") == "assistant":
+                speech_ts[ts_f] = speech_ts.get(ts_f, 0) + 1
+            if row.get("style") == "interjection":
+                interjection_ts[ts_f] = interjection_ts.get(ts_f, 0) + 1
+
+        for ts in interjection_ts:
+            if ts not in speech_ts:
+                report.add_error(f"ep={episode_index}: interjection at t={ts} has no paired speech atom")
+
+    def _check_plan_memory_consistency(
+        self,
+        persistent: Sequence[dict[str, Any]],
+        events: Sequence[dict[str, Any]],
+        report: ValidationReport,
+        episode_index: int,
+    ) -> None:
+        plan_ts = sorted({float(r["timestamp"]) for r in persistent if r.get("style") == "plan"})
+        memory_ts = sorted({float(r["timestamp"]) for r in persistent if r.get("style") == "memory"})
+        subtask_ts = sorted({float(r["timestamp"]) for r in persistent if r.get("style") == "subtask"})
+        interjection_ts = sorted(
+            {
+                float(r["timestamp"])
+                for r in events
+                if r.get("style") == "interjection" and r.get("timestamp") is not None
+            }
+        )
+
+        if persistent and not plan_ts:
+            report.add_warning(f"ep={episode_index}: persistent rows present but no plan emitted")
+        # every interjection should have a same-timestamp plan refresh
+        for ts in interjection_ts:
+            if ts not in set(plan_ts):
+                report.add_error(
+                    f"ep={episode_index}: interjection at t={ts} has no co-timestamped plan update"
+                )
+        # memory should be emitted at subtask boundaries (subset relation)
+        if memory_ts and subtask_ts:
+            mem_set = set(memory_ts)
+            sub_set = set(subtask_ts)
+            stray = sorted(mem_set - sub_set)
+            if stray:
+                report.add_warning(f"ep={episode_index}: memory rows at {stray} not at any subtask boundary")
+
+    def _check_vqa_json(
+        self,
+        events: Iterable[dict[str, Any]],
+        report: ValidationReport,
+        episode_index: int,
+    ) -> None:
+        for row in events:
+            if row.get("style") != "vqa" or row.get("role") != "assistant":
+                continue
+            content = row.get("content")
+            if content is None:
+                report.add_error(
+                    f"ep={episode_index}: VQA assistant row at t={row.get('timestamp')} has null content"
+                )
+                continue
+            try:
+                payload = json.loads(content)
+            except (TypeError, ValueError) as exc:
+                report.add_error(
+                    f"ep={episode_index}: VQA assistant content not valid JSON at t={row.get('timestamp')}: {exc}"
+                )
+                continue
+            shape = classify_vqa_answer(payload)
+            if shape is None:
+                report.add_error(
+                    f"ep={episode_index}: VQA assistant payload at t={row.get('timestamp')} does not match any known shape: keys={list(payload) if isinstance(payload, dict) else type(payload).__name__}"
+                )

src/lerobot/annotations/steerable_pipeline/vlm_client.py

@@ -0,0 +1,703 @@
+#!/usr/bin/env python
+
+# Copyright 2026 The HuggingFace Inc. team. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""Shared Qwen-VL client.
+
+The pipeline uses a single shared VLM across modules. vLLM is preferred when
+available (high throughput, JSON-guided decoding); transformers is the
+fallback. A ``stub`` backend is used for unit tests so fixtures never call
+into a real model.
+
+The client speaks one method, :meth:`VlmClient.generate_json`, which:
+
+- accepts a list of OpenAI/HF-style multimodal messages,
+- requests JSON output (``json_mode=True`` enables guided decoding when the
+  backend supports it),
+- batches requests transparently,
+- and reprompts once on a JSON parse failure with an inline correction
+  message before raising.
+"""
+
+from __future__ import annotations
+
+import atexit
+import base64
+import io
+import json
+import os
+import shlex
+import signal
+import subprocess
+import sys
+import threading
+import time
+import urllib.request
+from collections.abc import Callable, Sequence
+from concurrent.futures import ThreadPoolExecutor
+from dataclasses import dataclass
+from typing import Any, Protocol
+
+from .config import VlmConfig
+
+
+class VlmClient(Protocol):
+    """Protocol every backend must implement."""
+
+    def generate_json(
+        self,
+        messages_batch: Sequence[Sequence[dict[str, Any]]],
+        *,
+        max_new_tokens: int | None = None,
+        temperature: float | None = None,
+    ) -> list[Any]:
+        """Generate one JSON-decoded response per messages list."""
+
+
+@dataclass
+class StubVlmClient:
+    """Deterministic stub used in unit tests.
+
+    A test passes a callable that maps the *last user message text* (or, if
+    that is empty, the full message list) to a JSON-serializable response.
+    """
+
+    responder: Callable[[Sequence[dict[str, Any]]], Any]
+
+    def generate_json(
+        self,
+        messages_batch: Sequence[Sequence[dict[str, Any]]],
+        *,
+        max_new_tokens: int | None = None,
+        temperature: float | None = None,
+    ) -> list[Any]:
+        return [self.responder(list(messages)) for messages in messages_batch]
+
+
+def _strip_to_json(text: str) -> Any:
+    text = text.strip()
+    # Strip <think>...</think> blocks (Qwen3 Thinking style)
+    while "<think>" in text and "</think>" in text:
+        start = text.find("<think>")
+        end = text.find("</think>", start) + len("</think>")
+        text = (text[:start] + text[end:]).strip()
+    # Strip ```json ... ``` fences from chat-tuned backbones
+    if text.startswith("```"):
+        first = text.find("\n")
+        last = text.rfind("```")
+        if first != -1 and last != -1 and last > first:
+            text = text[first + 1 : last].strip()
+    try:
+        return json.loads(text)
+    except (ValueError, json.JSONDecodeError):
+        pass
+    # Fall back to extracting the first balanced {...} block.
+    obj_text = _extract_first_json_object(text)
+    if obj_text is None:
+        raise json.JSONDecodeError("No JSON object found", text, 0)
+    return json.loads(obj_text)
+
+
+def _extract_first_json_object(text: str) -> str | None:
+    """Return the first balanced ``{...}`` substring, ignoring braces in
+    string literals. Returns ``None`` if no balanced block is found."""
+    start = text.find("{")
+    if start < 0:
+        return None
+    depth = 0
+    in_string = False
+    escape = False
+    for i in range(start, len(text)):
+        ch = text[i]
+        if escape:
+            escape = False
+            continue
+        if ch == "\\":
+            escape = True
+            continue
+        # Note: ``escape`` is always False here — the ``if escape`` branch
+        # above already handled and reset it.
+        if ch == '"':
+            in_string = not in_string
+            continue
+        if in_string:
+            continue
+        if ch == "{":
+            depth += 1
+        elif ch == "}":
+            depth -= 1
+            if depth == 0:
+                return text[start : i + 1]
+    return None
+
+
+@dataclass
+class _GenericTextClient:
+    """Wraps any text-generation callable in JSON-mode + one-retry semantics."""
+
+    generate_text: Callable[[Sequence[Sequence[dict[str, Any]]], int, float], list[str]]
+    config: VlmConfig
+
+    def generate_json(
+        self,
+        messages_batch: Sequence[Sequence[dict[str, Any]]],
+        *,
+        max_new_tokens: int | None = None,
+        temperature: float | None = None,
+    ) -> list[Any]:
+        max_tok = max_new_tokens if max_new_tokens is not None else self.config.max_new_tokens
+        temp = temperature if temperature is not None else self.config.temperature
+        raw = self.generate_text(messages_batch, max_tok, temp)
+        out: list[Any] = []
+        for messages, text in zip(messages_batch, raw, strict=True):
+            try:
+                out.append(_strip_to_json(text))
+                continue
+            except (ValueError, json.JSONDecodeError):
+                pass
+            retry = list(messages) + [
+                {"role": "assistant", "content": text},
+                {
+                    "role": "user",
+                    "content": (
+                        "Your previous reply was not valid JSON. "
+                        "Reply with strictly valid JSON, no prose, no fences."
+                    ),
+                },
+            ]
+            retry_text = self.generate_text([retry], max_tok, temp)[0]
+            try:
+                out.append(_strip_to_json(retry_text))
+            except (ValueError, json.JSONDecodeError):
+                # After retry: log preview and return None instead of crashing
+                # the whole pipeline. Modules treat None as "skip".
+                preview = retry_text.strip().replace("\n", " ")[:200]
+                print(
+                    f"[vlm] WARNING: failed to parse JSON after retry; preview: {preview!r}",
+                    flush=True,
+                )
+                out.append(None)
+        return out
+
+
+def make_vlm_client(config: VlmConfig) -> VlmClient:
+    """Build the shared VLM client per the configured backend.
+
+    For ``stub``, callers should construct :class:`StubVlmClient` directly with
+    a responder callable. ``stub`` here is rejected to make accidental misuse
+    obvious.
+    """
+    if config.backend == "stub":
+        raise ValueError(
+            "Use StubVlmClient(...) directly for the stub backend; make_vlm_client builds real clients."
+        )
+    if config.backend == "vllm":
+        return _make_vllm_client(config)
+    if config.backend == "transformers":
+        return _make_transformers_client(config)
+    if config.backend == "openai":
+        return _make_openai_client(config)
+    raise ValueError(f"Unknown VLM backend: {config.backend!r}")
+
+
+def _make_vllm_client(config: VlmConfig) -> VlmClient:
+    try:
+        from vllm import LLM, SamplingParams  # type: ignore[import-not-found]
+    except ImportError as exc:
+        raise ImportError(
+            "vllm is required for backend='vllm'. Install with `pip install lerobot[annotations]`."
+        ) from exc
+    # Workaround for cuDNN 9.x + torch 2.8 conv3d regression that surfaces
+    # as CUDNN_STATUS_NOT_INITIALIZED in Qwen-VL vision-tower patch
+    # embedders. Setting LEROBOT_DISABLE_CUDNN=1 forces native PyTorch
+    # convolution kernels — slower but functional.
+    if os.environ.get("LEROBOT_DISABLE_CUDNN", "").lower() in {"1", "true", "yes"}:
+        import torch as _torch  # noqa: PLC0415  - optional GPU dep, deferred
+
+        _torch.backends.cudnn.enabled = False
+    llm_kwargs: dict[str, Any] = {
+        "model": config.model_id,
+        "tensor_parallel_size": config.tensor_parallel_size,
+        "gpu_memory_utilization": config.gpu_memory_utilization,
+        "trust_remote_code": config.trust_remote_code,
+    }
+    if config.max_model_len is not None:
+        llm_kwargs["max_model_len"] = config.max_model_len
+    llm = LLM(**llm_kwargs)
+
+    def _gen(batch: Sequence[Sequence[dict[str, Any]]], max_tok: int, temp: float) -> list[str]:
+        # ``guided_decoding`` would speed up parsing but its API differs across
+        # vllm releases (dict vs GuidedDecodingParams). The _GenericTextClient
+        # wrapper already has a one-retry JSON-recovery path, so we skip it.
+        params = SamplingParams(max_tokens=max_tok, temperature=temp)
+        # ``llm.chat`` handles chat-template application + multimodal input
+        # extraction (image/video blocks) internally, which ``llm.generate``
+        # does not.
+        outputs = llm.chat([list(m) for m in batch], params)
+        return [o.outputs[0].text for o in outputs]
+
+    return _GenericTextClient(_gen, config)
+
+
+def _make_transformers_client(config: VlmConfig) -> VlmClient:
+    try:
+        import torch  # type: ignore[import-not-found]
+        import transformers  # type: ignore[import-not-found]
+        from transformers import AutoProcessor  # type: ignore[import-not-found]
+    except ImportError as exc:
+        raise ImportError("transformers + torch are required for backend='transformers'.") from exc
+    auto_cls = getattr(transformers, "AutoModelForImageTextToText", None) or getattr(
+        transformers, "AutoModelForVision2Seq", None
+    )
+    if auto_cls is None:
+        raise ImportError(
+            "Neither AutoModelForImageTextToText nor AutoModelForVision2Seq is available in this "
+            "transformers version. Install transformers>=4.45 (which has AutoModelForImageTextToText) "
+            "for VL models."
+        )
+    processor = AutoProcessor.from_pretrained(config.model_id, trust_remote_code=config.trust_remote_code)
+    use_accelerate = os.environ.get("LEROBOT_TRANSFORMERS_DEVICE_MAP", "manual") != "manual"
+    # ``device_map='auto'`` triggers a known std::bad_alloc on the Qwen3-VL
+    # post-load dispatch path (the alloc fails in accelerate's hook setup
+    # even with TBs of host RAM). Default to manual: load on CPU with
+    # ``low_cpu_mem_usage=True``, then ``.to("cuda")``. Set
+    # ``LEROBOT_TRANSFORMERS_DEVICE_MAP=auto`` to opt back into the old path.
+    if use_accelerate:
+        model = auto_cls.from_pretrained(
+            config.model_id,
+            torch_dtype="auto",
+            device_map="auto",
+            low_cpu_mem_usage=True,
+            trust_remote_code=config.trust_remote_code,
+        )
+    else:
+        import torch as _torch  # noqa: PLC0415  - optional GPU dep, deferred
+
+        model = auto_cls.from_pretrained(
+            config.model_id,
+            torch_dtype=_torch.bfloat16,
+            low_cpu_mem_usage=True,
+            trust_remote_code=config.trust_remote_code,
+        )
+        model = model.to("cuda")
+    model.eval()
+
+    def _gen(batch: Sequence[Sequence[dict[str, Any]]], max_tok: int, temp: float) -> list[str]:
+        outs: list[str] = []
+        for messages in batch:
+            text = processor.apply_chat_template(messages, add_generation_prompt=True, tokenize=False)
+            inputs = processor(text=[text], return_tensors="pt").to(model.device)
+            with torch.no_grad():
+                gen = model.generate(
+                    **inputs,
+                    max_new_tokens=max_tok,
+                    temperature=temp,
+                    do_sample=temp > 0.0,
+                )
+            decoded = processor.batch_decode(
+                gen[:, inputs["input_ids"].shape[-1] :], skip_special_tokens=True
+            )[0]
+            outs.append(decoded)
+        return outs
+
+    return _GenericTextClient(_gen, config)
+
+
+def _make_openai_client(config: VlmConfig) -> VlmClient:
+    """Backend that talks to any OpenAI-compatible server.
+
+    Compatible with ``vllm serve``, ``transformers serve``,
+    ``ktransformers serve``, and hosted endpoints. By default the server
+    is expected to be already running. Set ``auto_serve=True`` to have
+    this client spawn one (default: ``transformers serve``), wait until
+    it's ready, and tear it down on process exit.
+
+    Image blocks ``{"type":"image", "image":<PIL.Image>}`` are
+    auto-converted to ``image_url`` data-URLs. Video blocks
+    ``{"type":"video", "video":[<PIL>...]}`` are forwarded as
+    multi-frame ``video_url`` items where supported.
+    """
+    try:
+        from openai import OpenAI  # type: ignore[import-not-found]
+    except ImportError as exc:
+        raise ImportError(
+            "openai package is required for backend='openai'. Install with `pip install openai`."
+        ) from exc
+
+    api_base = config.api_base
+    api_key = config.api_key
+    auto_serve = config.auto_serve
+    api_bases: list[str] = [api_base]
+
+    print(
+        f"[lerobot-annotate] backend=openai model={config.model_id} "
+        f"api_base={api_base} auto_serve={auto_serve}",
+        flush=True,
+    )
+    if auto_serve:
+        if config.parallel_servers > 1:
+            print(
+                f"[lerobot-annotate] spawning {config.parallel_servers} parallel servers",
+                flush=True,
+            )
+            api_bases = _spawn_parallel_inference_servers(config)
+        elif _server_is_up(api_base):
+            print(f"[lerobot-annotate] reusing server already up at {api_base}", flush=True)
+        else:
+            print("[lerobot-annotate] no server reachable; spawning one", flush=True)
+            api_base = _spawn_inference_server(config)
+            api_bases = [api_base]
+            print(f"[lerobot-annotate] server ready at {api_base}", flush=True)
+
+    clients = [OpenAI(base_url=base, api_key=api_key) for base in api_bases]
+    # round-robin counter for parallel mode
+    rr_counter = {"i": 0}
+
+    # ``mm_processor_kwargs`` is a vllm-specific extra; transformers serve
+    # rejects it with HTTP 422. Send it only when explicitly opted in via
+    # an env var (e.g. ``LEROBOT_OPENAI_SEND_MM_KWARGS=1`` for vllm).
+    send_mm_kwargs = os.environ.get("LEROBOT_OPENAI_SEND_MM_KWARGS", "").lower() in {"1", "true", "yes"}
+
+    rr_lock = threading.Lock()
+
+    def _one_call(messages: Sequence[dict[str, Any]], max_tok: int, temp: float) -> str:
+        api_messages, mm_kwargs = _to_openai_messages(messages)
+        kwargs: dict[str, Any] = {
+            "model": config.model_id,
+            "messages": api_messages,
+            "max_tokens": max_tok,
+            "temperature": temp,
+        }
+        extra_body: dict[str, Any] = {}
+        if send_mm_kwargs and mm_kwargs:
+            extra_body["mm_processor_kwargs"] = {**mm_kwargs, "do_sample_frames": True}
+        if config.chat_template_kwargs:
+            extra_body["chat_template_kwargs"] = config.chat_template_kwargs
+        if extra_body:
+            kwargs["extra_body"] = extra_body
+        with rr_lock:
+            chosen = clients[rr_counter["i"] % len(clients)]
+            rr_counter["i"] += 1
+        response = chosen.chat.completions.create(**kwargs)
+        return response.choices[0].message.content or ""
+
+    def _gen(batch: Sequence[Sequence[dict[str, Any]]], max_tok: int, temp: float) -> list[str]:
+        if len(batch) <= 1 or config.client_concurrency <= 1:
+            return [_one_call(messages, max_tok, temp) for messages in batch]
+        # Parallel fan-out — vllm batches these on the server side.
+        max_workers = min(config.client_concurrency, len(batch))
+        with ThreadPoolExecutor(max_workers=max_workers) as pool:
+            futures = [pool.submit(_one_call, messages, max_tok, temp) for messages in batch]
+            return [f.result() for f in futures]
+
+    return _GenericTextClient(_gen, config)
+
+
+def _spawn_parallel_inference_servers(config: VlmConfig) -> list[str]:
+    """Spawn ``config.parallel_servers`` independent vllm replicas.
+
+    Each replica:
+    - is pinned to a single GPU via ``CUDA_VISIBLE_DEVICES``
+    - listens on ``serve_port + i``
+    - is shut down via the same atexit hook as the single-server path
+
+    Returns the list of ``api_base`` URLs the client should round-robin
+    across.
+    """
+    n = config.parallel_servers
+    api_bases: list[str] = []
+    procs: list[subprocess.Popen] = []
+    ready_events: list[threading.Event] = []
+    # Multiple readiness signals — uvicorn's own banner is suppressed at
+    # ``--uvicorn-log-level warning``, so we also accept vllm's own
+    # "Starting vLLM API server" line and the route-listing line. The
+    # HTTP probe below is the ultimate fallback.
+    ready_markers = (
+        "Uvicorn running",
+        "Application startup complete",
+        "Starting vLLM API server",
+        "Available routes are",
+    )
+    # Single lock for all server-stream threads so multibyte chars from
+    # different servers don't interleave and tear UTF-8 sequences.
+    print_lock = threading.Lock()
+
+    base_cmd = config.serve_command or (
+        f"vllm serve {shlex.quote(config.model_id)} "
+        f"--tensor-parallel-size 1 "
+        f"--max-model-len {config.max_model_len or 32768} "
+        f"--uvicorn-log-level warning"
+    )
+
+    num_gpus = config.num_gpus if config.num_gpus > 0 else n
+    for i in range(n):
+        port = config.serve_port + i
+        gpu = i % num_gpus
+        env = os.environ.copy()
+        env["CUDA_VISIBLE_DEVICES"] = str(gpu)
+        cmd = base_cmd.replace("{port}", str(port)) if "{port}" in base_cmd else f"{base_cmd} --port {port}"
+        api_base = f"http://localhost:{port}/v1"
+        api_bases.append(api_base)
+        print(f"[server-{i}] launching on GPU {gpu} port {port}: {cmd}", flush=True)
+        proc = subprocess.Popen(
+            shlex.split(cmd),
+            stdout=subprocess.PIPE,
+            stderr=subprocess.STDOUT,
+            text=True,
+            bufsize=1,
+            env=env,
+        )
+        procs.append(proc)
+        ready = threading.Event()
+        ready_events.append(ready)
+
+        def _stream(idx: int, p: subprocess.Popen, ev: threading.Event) -> None:
+            # Read whole lines and emit each line atomically under the
+            # shared print_lock so output from N servers stays readable.
+            assert p.stdout is not None
+            for line in iter(p.stdout.readline, ""):
+                with print_lock:
+                    sys.stdout.write(f"[server-{idx}] {line}")
+                    if not line.endswith(("\n", "\r")):
+                        sys.stdout.write("\n")
+                    sys.stdout.flush()
+                if any(m in line for m in ready_markers):
+                    ev.set()
+
+        threading.Thread(target=_stream, args=(i, proc, ready), daemon=True).start()
+
+        def _probe(idx: int, base: str, ev: threading.Event, p: subprocess.Popen) -> None:
+            while not ev.is_set() and p.poll() is None:
+                if _server_is_up(base):
+                    print(f"[server-{idx}] ready (http probe)", flush=True)
+                    ev.set()
+                    return
+                time.sleep(2)
+
+        threading.Thread(target=_probe, args=(i, api_base, ready, proc), daemon=True).start()
+
+    def _shutdown() -> None:
+        for i, p in enumerate(procs):
+            if p.poll() is None:
+                print(f"[server-{i}] stopping pid={p.pid}", flush=True)
+                p.send_signal(signal.SIGINT)
+        for p in procs:
+            try:
+                p.wait(timeout=15)
+            except subprocess.TimeoutExpired:
+                p.kill()
+                p.wait(timeout=5)
+
+    atexit.register(_shutdown)
+
+    deadline = time.monotonic() + config.serve_ready_timeout_s
+    while any(not ev.is_set() for ev in ready_events) and time.monotonic() < deadline:
+        for i, p in enumerate(procs):
+            if p.poll() is not None:
+                raise RuntimeError(
+                    f"[server-{i}] inference server exited unexpectedly with rc={p.returncode}"
+                )
+        time.sleep(2)
+    if any(not ev.is_set() for ev in ready_events):
+        raise RuntimeError(f"[server] not all replicas became ready within {config.serve_ready_timeout_s}s")
+    print(f"[lerobot-annotate] all {n} servers ready: {api_bases}", flush=True)
+    return api_bases
+
+
+def _server_is_up(api_base: str) -> bool:
+    """Return True if ``api_base/models`` answers 200 within 2 seconds."""
+    url = api_base.rstrip("/") + "/models"
+    # ``api_base`` is the user-configured local-server URL we just spawned
+    # or the user passed in via ``--vlm.api_base``; the bandit B310 warning
+    # is for arbitrary user-controlled URLs with file:/ schemes which
+    # cannot reach this code path.
+    try:
+        with urllib.request.urlopen(url, timeout=2) as resp:  # noqa: S310  # nosec B310
+            return resp.status == 200
+    except Exception:  # noqa: BLE001
+        return False
+
+
+def _spawn_inference_server(config: VlmConfig) -> str:
+    """Spawn ``transformers serve`` (or ``serve_command``), wait until it
+    accepts ``/v1/models``, and register a shutdown hook.
+
+    Streams the server's stdout/stderr to the parent terminal in
+    real-time on a background thread so users can see model-load
+    progress and errors as they happen.
+
+    Returns the full ``api_base`` URL the OpenAI client should use.
+    """
+    cmd = config.serve_command
+    if not cmd:
+        cmd = (
+            f"transformers serve {shlex.quote(config.model_id)} "
+            f"--port {config.serve_port} --continuous-batching"
+        )
+    api_base = f"http://localhost:{config.serve_port}/v1"
+    print(f"[server] launching: {cmd}", flush=True)
+    proc = subprocess.Popen(
+        shlex.split(cmd),
+        stdout=subprocess.PIPE,
+        stderr=subprocess.STDOUT,
+        text=True,
+        bufsize=1,
+    )
+
+    # Watch the server output for the uvicorn readiness banner. This is
+    # more reliable than polling /v1/models because transformers serve
+    # rescans its cache on every model-list request, which can exceed
+    # the urllib timeout and trigger an infinite probe loop.
+    ready_event = threading.Event()
+    # See _spawn_parallel_inference_servers for why we accept these.
+    ready_markers = (
+        "Uvicorn running",
+        "Application startup complete",
+        "Starting vLLM API server",
+        "Available routes are",
+    )
+
+    def _probe() -> None:
+        while not ready_event.is_set() and proc.poll() is None:
+            if _server_is_up(api_base):
+                print("[server] ready (http probe)", flush=True)
+                ready_event.set()
+                return
+            time.sleep(2)
+
+    threading.Thread(target=_probe, daemon=True).start()
+
+    def _stream_output() -> None:
+        # Read raw chunks instead of iterating lines so tqdm progress
+        # bars (which overwrite using \r) flush in real time.
+        assert proc.stdout is not None
+        buf = ""
+        prefix_started = False
+        while True:
+            ch = proc.stdout.read(1)
+            if ch == "":
+                # process exited; flush any tail
+                if buf:
+                    sys.stdout.write(buf)
+                    sys.stdout.flush()
+                return
+            if not prefix_started:
+                sys.stdout.write("[server] ")
+                prefix_started = True
+            sys.stdout.write(ch)
+            sys.stdout.flush()
+            buf += ch
+            if ch in ("\n", "\r"):
+                if any(marker in buf for marker in ready_markers):
+                    ready_event.set()
+                buf = ""
+                prefix_started = False
+
+    threading.Thread(target=_stream_output, daemon=True).start()
+
+    def _shutdown() -> None:
+        if proc.poll() is None:
+            print(f"[server] stopping pid={proc.pid}", flush=True)
+            proc.send_signal(signal.SIGINT)
+            try:
+                proc.wait(timeout=15)
+            except subprocess.TimeoutExpired:
+                proc.kill()
+                proc.wait(timeout=5)
+
+    atexit.register(_shutdown)
+
+    deadline = time.monotonic() + config.serve_ready_timeout_s
+    while time.monotonic() < deadline:
+        if proc.poll() is not None:
+            raise RuntimeError(
+                f"[server] inference server exited unexpectedly with rc={proc.returncode}. "
+                f"See [server] log lines above for the cause."
+            )
+        if ready_event.wait(timeout=2):
+            return api_base
+    proc.terminate()
+    raise RuntimeError(f"[server] did not become ready within {config.serve_ready_timeout_s}s")
+
+
+def _to_openai_messages(
+    messages: Sequence[dict[str, Any]],
+) -> tuple[list[dict[str, Any]], dict[str, Any]]:
+    """Convert internal messages to OpenAI chat format.
+
+    Returns ``(api_messages, mm_kwargs)``. Multimodal-processor kwargs
+    (``fps`` from ``video_url`` blocks) are extracted out so the caller
+    can pass them via ``extra_body.mm_processor_kwargs`` rather than
+    inside the content blocks (which transformers serve rejects).
+
+    File-URL video blocks are inlined as base64 data URLs.
+    """
+    out_messages: list[dict[str, Any]] = []
+    mm_kwargs: dict[str, Any] = {}
+    for message in messages:
+        content = message.get("content")
+        if not isinstance(content, list):
+            out_messages.append({"role": message["role"], "content": content})
+            continue
+        out_blocks: list[dict[str, Any]] = []
+        for block in content:
+            block_type = block.get("type") if isinstance(block, dict) else None
+            if block_type == "text":
+                out_blocks.append({"type": "text", "text": block.get("text", "")})
+            elif block_type == "image":
+                out_blocks.append(
+                    {"type": "image_url", "image_url": {"url": _pil_to_data_url(block["image"])}}
+                )
+            elif block_type == "video":
+                frames = block.get("video", [])
+                for img in frames:
+                    out_blocks.append({"type": "image_url", "image_url": {"url": _pil_to_data_url(img)}})
+            elif block_type == "video_url":
+                video_url = dict(block["video_url"])
+                url = video_url.get("url", "")
+                if url.startswith("file://"):
+                    video_url["url"] = _file_to_data_url(url[len("file://") :])
+                out_blocks.append({"type": "video_url", "video_url": video_url})
+                fps = block.get("fps")
+                if fps is not None:
+                    mm_kwargs["fps"] = fps
+            else:
+                out_blocks.append(block)
+        out_messages.append({"role": message["role"], "content": out_blocks})
+    return out_messages, mm_kwargs
+
+
+def _file_to_data_url(path: str) -> str:
+    """Read a local video file and return a base64 ``data:video/mp4`` URL."""
+    with open(path, "rb") as f:
+        b64 = base64.b64encode(f.read()).decode("ascii")
+    return f"data:video/mp4;base64,{b64}"
+
+
+def _pil_to_data_url(image: Any) -> str:
+    """Encode a PIL.Image as a base64 data URL."""
+    buf = io.BytesIO()
+    image.save(buf, format="PNG")
+    b64 = base64.b64encode(buf.getvalue()).decode("ascii")
+    return f"data:image/png;base64,{b64}"
+
+
+def _messages_to_prompt(messages: Sequence[dict[str, Any]]) -> Any:
+    """Pass-through hook used by the vllm backend.
+
+    vllm exposes its own multimodal entry points that vary by version; for the
+    base flow we simply forward the raw message list and let the caller's
+    custom backend handle templating. Real deployments override this.
+    """
+    return list(messages)

src/lerobot/annotations/steerable_pipeline/writer.py

@@ -0,0 +1,356 @@
+#!/usr/bin/env python
+
+# Copyright 2026 The HuggingFace Inc. team. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""Final parquet rewrite.
+
+For every episode the writer:
+
+1. reads the staged module outputs,
+2. partitions them into a persistent slice (PERSISTENT_STYLES) and an event
+   slice (EVENT_ONLY_STYLES + style=None tool-call atoms),
+3. sorts each slice deterministically,
+4. broadcasts the persistent slice across every frame in the episode,
+5. for each frame, materializes the sublist of event rows whose timestamp
+   exactly equals that frame's timestamp,
+6. drops the legacy ``subtask_index`` column,
+7. writes the parquet shard back in place.
+
+The writer does NOT add a dataset-level ``tools`` column. Tool *calls* are
+emitted per-row via the existing ``tool_calls`` field on the v3.1 row
+struct for every speech atom. The tool *schema* (the description
+of the ``say`` function and its parameters) is a fixed code constant —
+``SAY_TOOL_SCHEMA`` below — and downstream chat-template consumers import
+it directly rather than reading a redundant per-row column.
+
+Invariants enforced here (and re-checked by the validator):
+
+- per-episode persistent slice is byte-identical across every frame;
+- ``language_events`` rows on a frame all have ``timestamp == frame_ts``
+  (timestamps come straight from the source parquet — never recomputed);
+- every row passes ``column_for_style(style)``.
+"""
+
+from __future__ import annotations
+
+import logging
+from collections import defaultdict
+from collections.abc import Iterable, Sequence
+from dataclasses import dataclass
+from pathlib import Path
+from typing import Any
+
+import pyarrow as pa
+import pyarrow.parquet as pq
+
+from lerobot.datasets.language import (
+    EVENT_ONLY_STYLES,
+    LANGUAGE_EVENTS,
+    LANGUAGE_PERSISTENT,
+    PERSISTENT_STYLES,
+    column_for_style,
+    validate_camera_field,
+)
+
+from .reader import EpisodeRecord
+from .staging import EpisodeStaging
+
+logger = logging.getLogger(__name__)
+
+
+# Tool schema constants live in lerobot.datasets.language — single
+# source of truth. Re-exported here so existing imports
+# (``from lerobot.annotations.steerable_pipeline.writer import SAY_TOOL_SCHEMA``)
+# keep working.
+from lerobot.datasets.language import DEFAULT_TOOLS, SAY_TOOL_SCHEMA  # noqa: F401, E402
+
+
+def _row_persistent_sort_key(row: dict[str, Any]) -> tuple:
+    return (float(row["timestamp"]), row.get("style") or "", row.get("role") or "")
+
+
+def _row_event_sort_key(row: dict[str, Any]) -> tuple:
+    # events are bucketed per-frame, but within a frame we still want determinism
+    return (
+        row.get("style") or "",
+        row.get("role") or "",
+        row.get("camera") or "",
+    )
+
+
+def _normalize_persistent_row(row: dict[str, Any]) -> dict[str, Any]:
+    """Coerce a staged row into the persistent column's struct shape."""
+    style = row.get("style")
+    if style not in PERSISTENT_STYLES:
+        raise ValueError(
+            f"persistent slice contains row with non-persistent style {style!r}; "
+            "row would be misrouted under column_for_style()"
+        )
+    if "timestamp" not in row:
+        raise ValueError(f"persistent row missing timestamp: {row!r}")
+    if "role" not in row:
+        # Surface a friendly error from the writer rather than letting
+        # the raw KeyError bubble out of the dict access below — modules
+        # are expected to always emit ``role``, but the validator
+        # currently doesn't check this so a future bug would otherwise
+        # be hard to triage.
+        raise ValueError(f"persistent row missing role: {row!r}")
+    camera = row.get("camera")
+    validate_camera_field(style, camera)
+    return {
+        "role": str(row["role"]),
+        "content": None if row.get("content") is None else str(row["content"]),
+        "style": style,
+        "timestamp": float(row["timestamp"]),
+        "camera": None if camera is None else str(camera),
+        "tool_calls": _normalize_tool_calls(row.get("tool_calls")),
+    }
+
+
+def _normalize_event_row(row: dict[str, Any]) -> dict[str, Any]:
+    """Coerce a staged row into the event column's struct shape (no timestamp)."""
+    style = row.get("style")
+    if style is not None and style not in EVENT_ONLY_STYLES:
+        raise ValueError(
+            f"event slice contains row with style {style!r}; expected None or one of {EVENT_ONLY_STYLES}"
+        )
+    if column_for_style(style) != LANGUAGE_EVENTS:
+        raise ValueError(f"event row with style {style!r} would not route to language_events")
+    if "role" not in row:
+        raise ValueError(f"event row missing role: {row!r}")
+    camera = row.get("camera")
+    validate_camera_field(style, camera)
+    return {
+        "role": str(row["role"]),
+        "content": None if row.get("content") is None else str(row["content"]),
+        "style": style,
+        "camera": None if camera is None else str(camera),
+        "tool_calls": _normalize_tool_calls(row.get("tool_calls")),
+    }
+
+
+def _normalize_tool_calls(value: Any) -> list[Any] | None:
+    if value is None:
+        return None
+    if not isinstance(value, list):
+        raise ValueError(f"tool_calls must be a list or None, got {type(value).__name__}")
+    return list(value)
+
+
+def _validate_atom_invariants(row: dict[str, Any]) -> None:
+    """At-least-one of content/tool_calls; style=None implies tool_calls."""
+    has_content = row.get("content") is not None
+    has_tools = row.get("tool_calls") is not None
+    if not (has_content or has_tools):
+        raise ValueError(f"row has neither content nor tool_calls: {row!r}")
+    if row.get("style") is None and not has_tools:
+        raise ValueError(f"style=None requires tool_calls: {row!r}")
+
+
+def _validate_speech_atom(row: dict[str, Any]) -> None:
+    """Speech atoms: role=assistant, style=None, content=None, say tool call."""
+    if row.get("style") is not None:
+        return  # not a speech atom
+    if row.get("role") != "assistant":
+        raise ValueError(f"speech atom must have role=assistant: {row!r}")
+    if row.get("content") is not None:
+        raise ValueError(f"speech atom must have content=null: {row!r}")
+    tool_calls = row.get("tool_calls")
+    if not tool_calls or not isinstance(tool_calls, list):
+        raise ValueError(f"speech atom must have non-empty tool_calls list: {row!r}")
+    first = tool_calls[0]
+    if not isinstance(first, dict):
+        raise ValueError(f"speech atom tool_calls[0] must be a dict: {row!r}")
+    if first.get("type") != "function":
+        raise ValueError(f"speech atom tool_calls[0].type must be 'function': {row!r}")
+    fn = first.get("function") or {}
+    if fn.get("name") != "say":
+        raise ValueError(f"speech atom tool_calls[0].function.name must be 'say': {row!r}")
+    args = fn.get("arguments") or {}
+    if not isinstance(args, dict) or "text" not in args or not isinstance(args["text"], str):
+        raise ValueError(f"speech atom must carry 'text' string in arguments: {row!r}")
+
+
+@dataclass
+class LanguageColumnsWriter:
+    """Rewrite ``data/chunk-*/file-*.parquet`` with the two language columns."""
+
+    drop_existing_subtask_index: bool = True
+
+    def write_all(
+        self,
+        records: Sequence[EpisodeRecord],
+        staging_dir: Path,
+        root: Path,
+    ) -> list[Path]:
+        episodes_by_path: dict[Path, list[EpisodeRecord]] = defaultdict(list)
+        for record in records:
+            episodes_by_path[record.data_path].append(record)
+
+        written: list[Path] = []
+        for path, eps in episodes_by_path.items():
+            self._rewrite_one(path, eps, staging_dir, root)
+            written.append(path)
+        return written
+
+    def _rewrite_one(
+        self,
+        path: Path,
+        episodes: Sequence[EpisodeRecord],
+        staging_dir: Path,
+        root: Path,
+    ) -> None:
+        table = pq.read_table(path)
+        n_rows = table.num_rows
+
+        # Ensure we cover every episode in the file. Episodes that don't have
+        # staging artifacts are passed through with empty annotation lists —
+        # this keeps the writer idempotent and safe for partial reruns.
+        staged_per_ep: dict[int, dict[str, list[dict[str, Any]]]] = {}
+        for record in episodes:
+            staging = EpisodeStaging(staging_dir, record.episode_index)
+            staged_per_ep[record.episode_index] = staging.read_all()
+
+        persistent_by_ep: dict[int, list[dict[str, Any]]] = {}
+        events_by_ep_ts: dict[int, dict[float, list[dict[str, Any]]]] = {}
+
+        for ep_index, ep_staged in staged_per_ep.items():
+            persistent_rows: list[dict[str, Any]] = []
+            event_rows: list[dict[str, Any]] = []  # carry timestamp until bucketed
+            for _module_name, rows in ep_staged.items():
+                for row in rows:
+                    style = row.get("style")
+                    if column_for_style(style) == LANGUAGE_PERSISTENT:
+                        persistent_rows.append(row)
+                    else:
+                        event_rows.append(row)
+
+            persistent_rows.sort(key=_row_persistent_sort_key)
+            normalized_persistent = []
+            for r in persistent_rows:
+                _validate_atom_invariants(r)
+                _validate_speech_atom(r)
+                normalized_persistent.append(_normalize_persistent_row(r))
+            persistent_by_ep[ep_index] = normalized_persistent
+
+            buckets: dict[float, list[dict[str, Any]]] = defaultdict(list)
+            for r in event_rows:
+                _validate_atom_invariants(r)
+                _validate_speech_atom(r)
+                ts = float(r["timestamp"])
+                buckets[ts].append(_normalize_event_row(r))
+            for ts in list(buckets.keys()):
+                buckets[ts].sort(key=_row_event_sort_key)
+            events_by_ep_ts[ep_index] = buckets
+
+        episode_col = (
+            table.column("episode_index").to_pylist() if "episode_index" in table.column_names else None
+        )
+        ts_col = table.column("timestamp").to_pylist() if "timestamp" in table.column_names else None
+        if episode_col is None or ts_col is None:
+            raise ValueError(f"{path} is missing 'episode_index' or 'timestamp' — required by the writer.")
+
+        per_row_persistent: list[list[dict[str, Any]]] = []
+        per_row_events: list[list[dict[str, Any]]] = []
+        for i in range(n_rows):
+            ep = episode_col[i]
+            ts = float(ts_col[i])
+            per_row_persistent.append(persistent_by_ep.get(ep, []))
+            buckets = events_by_ep_ts.get(ep, {})
+            per_row_events.append(buckets.get(ts, []))
+
+        new_table = self._materialize_table(
+            table, per_row_persistent, per_row_events, drop_old=self.drop_existing_subtask_index
+        )
+        # Atomic replace: write to a sibling tmp path and rename so a crash
+        # mid-write can't leave a half-written shard that ``pq.read_table``
+        # would then fail to open. ``Path.replace`` is atomic on POSIX +
+        # Windows when source and target sit on the same filesystem.
+        tmp_path = path.with_suffix(path.suffix + ".tmp")
+        pq.write_table(new_table, tmp_path)
+        tmp_path.replace(path)
+
+    def _materialize_table(
+        self,
+        table: pa.Table,
+        persistent: list[list[dict[str, Any]]],
+        events: list[list[dict[str, Any]]],
+        *,
+        drop_old: bool,
+    ) -> pa.Table:
+        cols = []
+        names = []
+        for name in table.column_names:
+            if drop_old and name == "subtask_index":
+                continue
+            if name in (LANGUAGE_PERSISTENT, LANGUAGE_EVENTS):
+                continue  # we'll re-add canonical versions
+            # Strip any legacy ``tools`` column previously emitted by older
+            # writers — the schema no longer uses it (constant lives in
+            # SAY_TOOL_SCHEMA / DEFAULT_TOOLS).
+            if name == "tools":
+                continue
+            cols.append(table.column(name))
+            names.append(name)
+
+        # We let pyarrow infer struct/list schema rather than passing the
+        # canonical type from `lerobot.datasets.language` directly: that type
+        # uses `pa.json_()` for the `tool_calls` element type, which
+        # `pa.array(..., type=...)` cannot materialize from Python lists on
+        # current pyarrow versions. The inferred schema round-trips through
+        # parquet and `LeRobotDataset` correctly — `tests/datasets/test_language.py`
+        # exercises the same flow.
+        persistent_arr = pa.array(persistent)
+        events_arr = pa.array(events)
+
+        cols.extend([persistent_arr, events_arr])
+        names.extend([LANGUAGE_PERSISTENT, LANGUAGE_EVENTS])
+
+        return pa.Table.from_arrays(cols, names=names)
+
+
+def speech_atom(timestamp: float, text: str) -> dict[str, Any]:
+    """Build a canonical speech tool-call atom for the events column."""
+    return {
+        "role": "assistant",
+        "content": None,
+        "style": None,
+        "timestamp": float(timestamp),
+        "camera": None,
+        "tool_calls": [
+            {
+                "type": "function",
+                "function": {
+                    "name": "say",
+                    "arguments": {"text": text},
+                },
+            }
+        ],
+    }
+
+
+def normalize_rows_for_writer(
+    rows: Iterable[dict[str, Any]],
+) -> tuple[list[dict[str, Any]], list[dict[str, Any]]]:
+    """Helper used by tests/validators to partition a flat row list into
+    (persistent_rows, event_rows) using ``column_for_style``.
+    """
+    persistent: list[dict[str, Any]] = []
+    events: list[dict[str, Any]] = []
+    for row in rows:
+        if column_for_style(row.get("style")) == LANGUAGE_PERSISTENT:
+            persistent.append(row)
+        else:
+            events.append(row)
+    return persistent, events

src/lerobot/cameras/opencv/camera_opencv.py

@@ -199,12 +199,13 @@ class OpenCVCamera(Camera):
             DeviceNotConnectedError: If the camera is not connected.
         """
 
-        # Set FOURCC first (if specified) as it can affect available FPS/resolution options
-        if self.config.fourcc is not None:
-            self._validate_fourcc()
         if self.videocapture is None:
             raise DeviceNotConnectedError(f"{self} videocapture is not initialized")
 
+        set_fourcc_after_size_and_fps = platform.system() == "Windows"
+        if self.config.fourcc is not None and not set_fourcc_after_size_and_fps:
+            self._validate_fourcc()
+
         default_width = int(round(self.videocapture.get(cv2.CAP_PROP_FRAME_WIDTH)))
         default_height = int(round(self.videocapture.get(cv2.CAP_PROP_FRAME_HEIGHT)))
 
@@ -222,6 +223,11 @@ class OpenCVCamera(Camera):
         else:
             self._validate_fps()
 
+        if self.config.fourcc is not None and set_fourcc_after_size_and_fps:
+            # On Windows with DSHOW, changing the resolution can silently override the FOURCC setting.
+            # Set FOURCC last to make sure the requested pixel format is actually enforced.
+            self._validate_fourcc()
+
     def _validate_fps(self) -> None:
         """Validates and sets the camera's frames per second (FPS)."""
 

src/lerobot/cameras/realsense/camera_realsense.py

@@ -17,6 +17,7 @@ Provides the RealSenseCamera class for capturing frames from Intel RealSense cam
 """
 
 import logging
+import sys
 import time
 from threading import Event, Lock, Thread
 from typing import TYPE_CHECKING, Any
@@ -41,6 +42,7 @@ from ..utils import get_cv2_rotation
 from .configuration_realsense import RealSenseCameraConfig
 
 logger = logging.getLogger(__name__)
+pkg_name = "pyrealsense2-macosx" if sys.platform == "darwin" else "pyrealsense2"
 
 
 class RealSenseCamera(Camera):
@@ -114,7 +116,7 @@ class RealSenseCamera(Camera):
         Args:
             config: The configuration settings for the camera.
         """
-        require_package("pyrealsense2", extra="intelrealsense")
+        require_package(pkg_name, extra="intelrealsense", import_name="pyrealsense2")
         super().__init__(config)
 
         self.config = config

src/lerobot/common/train_utils.py

@@ -99,6 +99,7 @@ def save_checkpoint(
         optimizer (Optimizer | None, optional): The optimizer to save the state from. Defaults to None.
         scheduler (LRScheduler | None, optional): The scheduler to save the state from. Defaults to None.
         preprocessor: The preprocessor/pipeline to save. Defaults to None.
+        postprocessor: The postprocessor/pipeline to save. Defaults to None.
     """
     pretrained_dir = checkpoint_dir / PRETRAINED_MODEL_DIR
     policy.save_pretrained(pretrained_dir)

src/lerobot/common/wandb_utils.py

@@ -41,8 +41,12 @@ def cfg_to_group(
             return tag
         return tag[:max_tag_length]
 
+    if cfg.is_reward_model_training:
+        trainable_tag = f"reward_model:{cfg.reward_model.type}"
+    else:
+        trainable_tag = f"policy:{cfg.policy.type}"
     lst = [
-        f"policy:{cfg.policy.type}",
+        trainable_tag,
         f"seed:{cfg.seed}",
     ]
     if cfg.dataset is not None:

src/lerobot/configs/__init__.py

@@ -24,6 +24,7 @@ Import them directly: ``from lerobot.configs.train import TrainPipelineConfig``
 from .dataset import DatasetRecordConfig
 from .default import DatasetConfig, EvalConfig, PeftConfig, WandBConfig
 from .policies import PreTrainedConfig
+from .recipe import MessageTurn, TrainingRecipe, load_recipe
 from .types import (
     FeatureType,
     NormalizationMode,
@@ -31,6 +32,12 @@ from .types import (
     PolicyFeature,
     RTCAttentionSchedule,
 )
+from .video import (
+    VALID_VIDEO_CODECS,
+    VIDEO_ENCODER_INFO_KEYS,
+    VideoEncoderConfig,
+    camera_encoder_defaults,
+)
 
 __all__ = [
     # Types
@@ -43,7 +50,16 @@ __all__ = [
     "DatasetRecordConfig",
     "DatasetConfig",
     "EvalConfig",
+    "MessageTurn",
     "PeftConfig",
     "PreTrainedConfig",
+    "TrainingRecipe",
     "WandBConfig",
+    "load_recipe",
+    "VideoEncoderConfig",
+    # Defaults
+    "camera_encoder_defaults",
+    # Constants
+    "VALID_VIDEO_CODECS",
+    "VIDEO_ENCODER_INFO_KEYS",
 ]

src/lerobot/configs/dataset.py

@@ -14,10 +14,12 @@
 
 """Shared dataset recording configuration used by both ``lerobot-record`` and ``lerobot-rollout``."""
 
-from dataclasses import dataclass
+from dataclasses import dataclass, field
 from datetime import datetime
 from pathlib import Path
 
+from .video import VideoEncoderConfig, camera_encoder_defaults
+
 
 @dataclass
 class DatasetRecordConfig:
@@ -55,10 +57,9 @@ class DatasetRecordConfig:
     # Number of episodes to record before batch encoding videos
     # Set to 1 for immediate encoding (default behavior), or higher for batched encoding
     video_encoding_batch_size: int = 1
-    # Video codec for encoding videos. Options: 'h264', 'hevc', 'libsvtav1', 'auto',
-    # or hardware-specific: 'h264_videotoolbox', 'h264_nvenc', 'h264_vaapi', 'h264_qsv'.
-    # Use 'auto' to auto-detect the best available hardware encoder.
-    vcodec: str = "libsvtav1"
+    # Video encoder settings for camera MP4s (codec, quality, GOP, etc.). Tuned via CLI nested keys,
+    # e.g. ``--dataset.camera_encoder.vcodec=h264`` (see ``VideoEncoderConfig``).
+    camera_encoder: VideoEncoderConfig = field(default_factory=camera_encoder_defaults)
     # Enable streaming video encoding: encode frames in real-time during capture instead
     # of writing PNG images first. Makes save_episode() near-instant. More info in the documentation: https://huggingface.co/docs/lerobot/streaming_video_encoding
     streaming_encoding: bool = False

src/lerobot/configs/default.py

@@ -17,7 +17,7 @@
 from dataclasses import dataclass, field
 
 from lerobot.transforms import ImageTransformsConfig
-from lerobot.utils.import_utils import get_safe_default_codec
+from lerobot.utils.import_utils import get_safe_default_video_backend
 
 
 @dataclass
@@ -34,7 +34,7 @@ class DatasetConfig:
     image_transforms: ImageTransformsConfig = field(default_factory=ImageTransformsConfig)
     revision: str | None = None
     use_imagenet_stats: bool = True
-    video_backend: str = field(default_factory=get_safe_default_codec)
+    video_backend: str = field(default_factory=get_safe_default_video_backend)
     # When True, video frames are returned as uint8 tensors (0-255) instead of float32 (0.0-1.0).
     # This reduces memory and speeds up DataLoader IPC. The training pipeline handles the conversion.
     return_uint8: bool = False
@@ -117,3 +117,9 @@ class PeftConfig:
     # the rank used for the adapter. In general a higher rank means more trainable parameters and closer to full
     # fine-tuning.
     r: int = 16
+
+    # Alpha parameter for LoRA scaling (scaling = lora_alpha / r).
+    # In general, a higher alpha means stronger adaptation signal.
+    # If None, the PEFT library defaults to alpha=8, which may dampen high-rank adapters.
+    # Common values are r (alpha == rank) or 2*r.
+    lora_alpha: int | None = None

src/lerobot/configs/eval.py

@@ -18,8 +18,8 @@ from logging import getLogger
 from pathlib import Path
 
 from lerobot import envs, policies  # noqa: F401
-from lerobot.configs import parser
 
+from . import parser
 from .default import EvalConfig
 from .policies import PreTrainedConfig
 
@@ -46,8 +46,11 @@ class EvalPipelineConfig:
         # HACK: We parse again the cli args here to get the pretrained path if there was one.
         policy_path = parser.get_path_arg("policy")
         if policy_path:
-            cli_overrides = parser.get_cli_overrides("policy")
-            self.policy = PreTrainedConfig.from_pretrained(policy_path, cli_overrides=cli_overrides)
+            yaml_overrides = parser.get_yaml_overrides("policy")
+            cli_overrides = parser.get_cli_overrides("policy") or []
+            self.policy = PreTrainedConfig.from_pretrained(
+                policy_path, cli_overrides=yaml_overrides + cli_overrides
+            )
             self.policy.pretrained_path = Path(policy_path)
 
         else:

src/lerobot/configs/parser.py

@@ -13,8 +13,10 @@
 # limitations under the License.
 import importlib
 import inspect
+import json
 import pkgutil
 import sys
+import tempfile
 from argparse import ArgumentError
 from collections.abc import Callable, Iterable, Sequence
 from functools import wraps
@@ -24,6 +26,7 @@ from types import ModuleType
 from typing import Any, TypeVar, cast
 
 import draccus
+import yaml  # type: ignore[import-untyped]
 
 from lerobot.utils.utils import has_method
 
@@ -32,6 +35,29 @@ F = TypeVar("F", bound=Callable[..., object])
 PATH_KEY = "path"
 PLUGIN_DISCOVERY_SUFFIX = "discover_packages_path"
 
+# Storage for path args extracted from YAML/JSON config files, so that
+# get_path_arg() can find them even when they weren't passed via CLI.
+_config_path_args: dict[str, str] = {}
+
+# Storage for non-path YAML overrides so validate() can pass them to from_pretrained.
+_config_yaml_overrides: dict[str, list[str]] = {}
+
+
+def _flatten_to_cli_args(d: dict, prefix: str = "") -> list[str]:
+    """Recursively flatten a nested dict to CLI-style args (e.g. {"lr": 1e-4} -> ["--lr=0.0001"])."""
+    args = []
+    for key, value in d.items():
+        if key in (PATH_KEY, draccus.CHOICE_TYPE_KEY):
+            continue
+        full_key = f"{prefix}.{key}" if prefix else key
+        if isinstance(value, bool):
+            value = str(value).lower()
+        if isinstance(value, dict):
+            args.extend(_flatten_to_cli_args(value, full_key))
+        elif value is not None and not isinstance(value, list):
+            args.append(f"--{full_key}={value}")
+    return args
+
 
 def get_cli_overrides(field_name: str, args: Sequence[str] | None = None) -> list[str] | None:
     """Parses arguments from cli at a given nested attribute level.
@@ -145,7 +171,14 @@ def load_plugin(plugin_path: str) -> None:
 
 
 def get_path_arg(field_name: str, args: Sequence[str] | None = None) -> str | None:
-    return parse_arg(f"{field_name}.{PATH_KEY}", args)
+    result = parse_arg(f"{field_name}.{PATH_KEY}", args)
+    if result is None:
+        result = _config_path_args.get(field_name)
+    return result
+
+
+def get_yaml_overrides(field_name: str) -> list[str]:
+    return _config_yaml_overrides.get(field_name, [])
 
 
 def get_type_arg(field_name: str, args: Sequence[str] | None = None) -> str | None:
@@ -192,6 +225,51 @@ def filter_path_args(fields_to_filter: str | list[str], args: Sequence[str] | No
     return filtered_args
 
 
+def extract_path_fields_from_config(config_path: str, path_fields: list[str]) -> str:
+    """Extract `path` fields from a YAML/JSON config before draccus processes it.
+
+    When a user specifies e.g. ``policy.path: lerobot/smolvla_base`` in a YAML config,
+    draccus will fail because ``path`` is not a valid field on policy config classes.
+    This function extracts those path values, stores them in ``_config_path_args`` for
+    later retrieval by ``get_path_arg()``, and returns a cleaned temp config file path.
+    """
+    config_file = Path(config_path)
+    suffix = config_file.suffix.lower()
+
+    if suffix in (".yaml", ".yml"):
+        with open(config_file) as f:
+            config_data = yaml.safe_load(f)
+    elif suffix == ".json":
+        with open(config_file) as f:
+            config_data = json.load(f)
+    else:
+        return config_path
+
+    if not isinstance(config_data, dict):
+        return config_path
+
+    modified = False
+    for field in path_fields:
+        if field in config_data and isinstance(config_data[field], dict) and PATH_KEY in config_data[field]:
+            _config_path_args[field] = str(config_data[field].pop(PATH_KEY))
+            remaining = config_data[field]
+            if remaining:
+                _config_yaml_overrides[field] = _flatten_to_cli_args(remaining)
+            del config_data[field]
+            modified = True
+
+    if not modified:
+        return config_path
+
+    # Write cleaned config to a temp file
+    with tempfile.NamedTemporaryFile(mode="w", suffix=suffix, delete=False) as tmp:
+        if suffix in (".yaml", ".yml"):
+            yaml.dump(config_data, tmp, default_flow_style=False)
+        else:
+            json.dump(config_data, tmp, indent=2)
+    return tmp.name
+
+
 def wrap(config_path: Path | None = None) -> Callable[[F], F]:
     """
     HACK: Similar to draccus.wrap but does three additional things:
@@ -225,11 +303,20 @@ def wrap(config_path: Path | None = None) -> Callable[[F], F]:
                 if has_method(argtype, "__get_path_fields__"):
                     path_fields = argtype.__get_path_fields__()
                     cli_args = filter_path_args(path_fields, cli_args)
+                    # Also extract path fields from the YAML/JSON config file
+                    if config_path_cli:
+                        config_path_cli = extract_path_fields_from_config(config_path_cli, path_fields)
                 if has_method(argtype, "from_pretrained") and config_path_cli:
                     cli_args = filter_arg("config_path", cli_args)
                     cfg = argtype.from_pretrained(config_path_cli, cli_args=cli_args)
                 else:
-                    cfg = draccus.parse(config_class=argtype, config_path=config_path, args=cli_args)
+                    if config_path_cli:
+                        cli_args = filter_arg("config_path", cli_args)
+                    cfg = draccus.parse(
+                        config_class=argtype,
+                        config_path=config_path_cli or config_path,
+                        args=cli_args,
+                    )
             response = fn(cfg, *args, **kwargs)
             return response
 

src/lerobot/configs/recipe.py

@@ -0,0 +1,206 @@
+#!/usr/bin/env python
+
+# Copyright 2026 The HuggingFace Inc. team. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from __future__ import annotations
+
+import re
+from dataclasses import dataclass
+from pathlib import Path
+from typing import Any, Literal, get_args
+
+MessageRole = Literal["user", "assistant", "system", "tool"]
+MessageStream = Literal["high_level", "low_level"]
+
+DEFAULT_BINDINGS = {
+    "subtask": "active_at(t, style=subtask)",
+    "memory": "active_at(t, style=memory)",
+    "plan": "active_at(t, style=plan)",
+    "speech": "emitted_at(t, role=assistant, tool_name=say)",
+    "interjection": "emitted_at(t, style=interjection)",
+    "vqa": "emitted_at(t, style=vqa, role=assistant)",
+    "vqa_query": "emitted_at(t, style=vqa, role=user)",
+}
+
+PLACEHOLDER_RE = re.compile(r"\$\{([A-Za-z_][A-Za-z0-9_]*)\}")
+"""``${name}`` placeholder pattern used by both recipe binding-reference
+discovery (here) and rendered-message substitution (in ``language_render``)."""
+
+_VALID_ROLES = frozenset(get_args(MessageRole))
+_VALID_STREAMS = frozenset(get_args(MessageStream))
+
+
+@dataclass
+class MessageTurn:
+    """A single chat-style turn in a recipe template.
+
+    ``content`` may be a plain string, a list of HF-style multimodal blocks, or
+    ``None`` when ``tool_calls_from`` supplies tool-call payloads instead.
+    ``stream`` tags the turn for downstream filtering, ``target`` flags it as a
+    training target, and ``if_present`` skips the turn when the named binding
+    resolves to ``None``.
+    """
+
+    role: MessageRole
+    content: str | list[dict[str, Any]] | None = None
+    stream: MessageStream | None = None
+    target: bool = False
+    if_present: str | None = None
+    tool_calls_from: str | None = None
+
+    def __post_init__(self) -> None:
+        """Validate role, stream, and content after dataclass construction."""
+        if self.role not in _VALID_ROLES:
+            raise ValueError(f"Unsupported message role: {self.role!r}")
+        # ``stream`` is typed Optional only so the dataclass can keep its
+        # field ordering, but recipes must always tag every turn with a
+        # stream — the renderer's ``_validate_rendered`` would reject
+        # ``None`` later on. Fail at construction so the bad recipe is
+        # caught at YAML load time rather than at the first sample.
+        if self.stream is None:
+            raise ValueError(
+                f"MessageTurn(role={self.role!r}) is missing a stream — "
+                f"every turn must declare one of {sorted(_VALID_STREAMS)}."
+            )
+        if self.stream not in _VALID_STREAMS:
+            raise ValueError(f"Unsupported message stream: {self.stream!r}")
+        if self.content is None and self.tool_calls_from is None:
+            raise ValueError("MessageTurn.content is required unless tool_calls_from is set.")
+        if self.content is not None and not isinstance(self.content, (str, list)):
+            raise TypeError("MessageTurn.content must be a string, a list of HF-style blocks, or None.")
+        if isinstance(self.content, list):
+            for block in self.content:
+                if not isinstance(block, dict) or "type" not in block:
+                    raise ValueError(
+                        "Multimodal content blocks must be HF-style dictionaries with a type key."
+                    )
+
+    @classmethod
+    def from_dict(cls, data: dict[str, Any]) -> MessageTurn:
+        """Construct a :class:`MessageTurn` from a plain dictionary."""
+        return cls(**data)
+
+
+@dataclass
+class TrainingRecipe:
+    """A recipe describing how to render training samples from language rows.
+
+    A recipe is either a *message recipe* (``messages`` plus optional
+    ``bindings``) or a *blend recipe* (``blend`` mapping names to weighted
+    sub-recipes). ``weight`` is only meaningful inside a blend.
+    """
+
+    messages: list[MessageTurn] | None = None
+    bindings: dict[str, str] | None = None
+    blend: dict[str, TrainingRecipe] | None = None
+    weight: float | None = None
+
+    def __post_init__(self) -> None:
+        """Validate that exactly one of ``messages`` or ``blend`` is set."""
+        if self.messages is not None and self.blend is not None:
+            raise ValueError("TrainingRecipe must set only one of messages or blend.")
+        if self.messages is None and self.blend is None:
+            raise ValueError("TrainingRecipe must set one of messages or blend.")
+
+        if self.messages is not None:
+            self._validate_message_recipe()
+        if self.blend is not None:
+            self._validate_blend_recipe()
+
+    @classmethod
+    def from_dict(cls, data: dict[str, Any]) -> TrainingRecipe:
+        """Construct a :class:`TrainingRecipe` from a nested dictionary."""
+        data = dict(data)
+        if data.get("messages") is not None:
+            data["messages"] = [
+                turn if isinstance(turn, MessageTurn) else MessageTurn.from_dict(turn)
+                for turn in data["messages"]
+            ]
+        if data.get("blend") is not None:
+            data["blend"] = {
+                name: recipe if isinstance(recipe, TrainingRecipe) else cls.from_dict(recipe)
+                for name, recipe in data["blend"].items()
+            }
+        return cls(**data)
+
+    @classmethod
+    def from_yaml(cls, path: str | Path) -> TrainingRecipe:
+        """Load a :class:`TrainingRecipe` from a YAML file at ``path``."""
+        import yaml  # type: ignore[import-untyped]
+
+        with open(path) as f:
+            data = yaml.safe_load(f)
+        if not isinstance(data, dict):
+            raise ValueError(f"Recipe YAML must contain a mapping at the top level: {path}")
+        return cls.from_dict(data)
+
+    def _validate_message_recipe(self) -> None:
+        """Ensure every templated binding is known and at least one turn is a target."""
+        assert self.messages is not None
+        known_bindings = set(DEFAULT_BINDINGS) | set(self.bindings or {}) | {"task"}
+
+        for turn in self.messages:
+            missing = self._referenced_bindings(turn) - known_bindings
+            if missing:
+                raise ValueError(f"MessageTurn references unknown binding(s): {sorted(missing)}")
+
+        if not any(turn.target for turn in self.messages):
+            raise ValueError("Message recipes must contain at least one target turn.")
+
+    def _validate_blend_recipe(self) -> None:
+        """Ensure each blend component is a non-empty, weighted message recipe."""
+        assert self.blend is not None
+        if not self.blend:
+            raise ValueError("Blend recipes must contain at least one component.")
+
+        for name, recipe in self.blend.items():
+            if recipe.blend is not None:
+                raise ValueError(f"Blend component {name!r} cannot itself define a blend.")
+            if recipe.messages is None:
+                raise ValueError(f"Blend component {name!r} must define messages.")
+            if recipe.weight is None:
+                raise ValueError(f"Blend component {name!r} must define weight.")
+            if recipe.weight <= 0:
+                raise ValueError(f"Blend component {name!r} must have a positive weight.")
+
+    def _referenced_bindings(self, turn: MessageTurn) -> set[str]:
+        """Return the binding names that ``turn`` references via placeholders or attributes."""
+        names: set[str] = set()
+        if turn.if_present is not None:
+            names.add(turn.if_present)
+        if turn.tool_calls_from is not None:
+            names.add(turn.tool_calls_from)
+        names.update(_placeholders_in_content(turn.content))
+        return names
+
+
+def _placeholders_in_content(content: str | list[dict[str, Any]] | None) -> set[str]:
+    """Return the set of ``${name}`` placeholders found anywhere in ``content``."""
+    if content is None:
+        return set()
+    if isinstance(content, str):
+        return set(PLACEHOLDER_RE.findall(content))
+
+    names: set[str] = set()
+    for block in content:
+        for value in block.values():
+            if isinstance(value, str):
+                names.update(PLACEHOLDER_RE.findall(value))
+    return names
+
+
+def load_recipe(path: str | Path) -> TrainingRecipe:
+    """Load a :class:`TrainingRecipe` from a YAML file at ``path``."""
+    return TrainingRecipe.from_yaml(path)

src/lerobot/configs/rewards.py

@@ -0,0 +1,164 @@
+# Copyright 2026 The HuggingFace Inc. team. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import abc
+import builtins
+import json
+import logging
+import os
+import tempfile
+from dataclasses import dataclass, field
+from pathlib import Path
+from typing import Any, TypeVar
+
+import draccus
+from huggingface_hub import hf_hub_download
+from huggingface_hub.constants import CONFIG_NAME
+from huggingface_hub.errors import HfHubHTTPError
+
+from lerobot.optim.optimizers import OptimizerConfig
+from lerobot.optim.schedulers import LRSchedulerConfig
+from lerobot.utils.device_utils import auto_select_torch_device, is_torch_device_available
+from lerobot.utils.hub import HubMixin
+
+from .types import PolicyFeature
+
+T = TypeVar("T", bound="RewardModelConfig")
+logger = logging.getLogger(__name__)
+
+
+@dataclass
+class RewardModelConfig(draccus.ChoiceRegistry, HubMixin, abc.ABC):
+    """Base configuration for reward models.
+
+    Args:
+    input_features: A dictionary defining the PolicyFeature of the input data for the reward. The key represents
+        the input data name, and the value is PolicyFeature, which consists of FeatureType and shape attributes.
+    output_features: A dictionary defining the PolicyFeature of the output data for the reward. The key represents
+        the output data name, and the value is PolicyFeature, which consists of FeatureType and shape attributes.
+    """
+
+    # Reuses PolicyFeature
+    input_features: dict[str, PolicyFeature] = field(default_factory=dict)
+    output_features: dict[str, PolicyFeature] = field(default_factory=dict)
+
+    device: str | None = None
+
+    pretrained_path: str | None = None
+
+    push_to_hub: bool = False
+    repo_id: str | None = None
+
+    # Hub metadata
+    license: str | None = None
+    tags: list[str] | None = None
+    private: bool | None = None
+
+    def __post_init__(self) -> None:
+        if not self.device or not is_torch_device_available(self.device):
+            auto_device = auto_select_torch_device()
+            logger.warning(f"Device '{self.device}' is not available. Switching to '{auto_device}'.")
+            self.device = auto_device.type
+
+    @property
+    def type(self) -> str:
+        choice_name = self.get_choice_name(self.__class__)
+        if not isinstance(choice_name, str):
+            raise TypeError(f"Expected string from get_choice_name, got {type(choice_name)}")
+        return choice_name
+
+    @property
+    def observation_delta_indices(self) -> list | None:  # type: ignore[type-arg]
+        return None
+
+    @property
+    def action_delta_indices(self) -> list | None:  # type: ignore[type-arg]
+        return None
+
+    @property
+    def reward_delta_indices(self) -> list | None:  # type: ignore[type-arg]
+        return None
+
+    def get_optimizer_preset(self) -> OptimizerConfig | None:
+        """Default optimizer for this reward model, or ``None`` for zero-shot models."""
+        return None
+
+    def get_scheduler_preset(self) -> LRSchedulerConfig | None:
+        return None
+
+    def validate_features(self) -> None:
+        pass
+
+    def _save_pretrained(self, save_directory: Path) -> None:
+        with open(save_directory / CONFIG_NAME, "w") as f, draccus.config_type("json"):
+            draccus.dump(self, f, indent=4)
+
+    @classmethod
+    def from_pretrained(
+        cls: builtins.type[T],
+        pretrained_name_or_path: str | Path,
+        *,
+        force_download: bool = False,
+        resume_download: bool | None = None,
+        proxies: dict[Any, Any] | None = None,
+        token: str | bool | None = None,
+        cache_dir: str | Path | None = None,
+        local_files_only: bool = False,
+        revision: str | None = None,
+        **reward_kwargs: Any,
+    ) -> T:
+        model_id = str(pretrained_name_or_path)
+        config_file: str | None = None
+        if Path(model_id).is_dir():
+            if CONFIG_NAME in os.listdir(model_id):
+                config_file = os.path.join(model_id, CONFIG_NAME)
+            else:
+                logger.error(f"{CONFIG_NAME} not found in {Path(model_id).resolve()}")
+        else:
+            try:
+                config_file = hf_hub_download(
+                    repo_id=model_id,
+                    filename=CONFIG_NAME,
+                    revision=revision,
+                    cache_dir=cache_dir,
+                    force_download=force_download,
+                    proxies=proxies,
+                    resume_download=resume_download,
+                    token=token,
+                    local_files_only=local_files_only,
+                )
+            except HfHubHTTPError as e:
+                raise FileNotFoundError(
+                    f"{CONFIG_NAME} not found on the HuggingFace Hub in {model_id}"
+                ) from e
+
+        if config_file is None:
+            raise FileNotFoundError(f"{CONFIG_NAME} not found in {model_id}")
+
+        # HACK: Parse the original config to get the config subclass, so that we can
+        # apply cli overrides.
+        with draccus.config_type("json"):
+            orig_config = draccus.parse(cls, config_file, args=[])
+
+        with open(config_file) as f:
+            config = json.load(f)
+
+        config.pop("type", None)
+        with tempfile.NamedTemporaryFile("w+", delete=False, suffix=".json") as f:
+            json.dump(config, f)
+            config_file = f.name
+
+        cli_overrides = reward_kwargs.pop("cli_overrides", [])
+        with draccus.config_type("json"):
+            return draccus.parse(orig_config.__class__, config_file, args=cli_overrides)

src/lerobot/configs/train.py

@@ -13,7 +13,9 @@
 # limitations under the License.
 import builtins
 import datetime as dt
+import json
 import os
+import tempfile
 from dataclasses import dataclass, field
 from pathlib import Path
 from typing import Any
@@ -23,21 +25,60 @@ from huggingface_hub import hf_hub_download
 from huggingface_hub.errors import HfHubHTTPError
 
 from lerobot import envs
-from lerobot.configs import parser
 from lerobot.optim import LRSchedulerConfig, OptimizerConfig
 from lerobot.utils.hub import HubMixin
+from lerobot.utils.sample_weighting import SampleWeightingConfig
 
+from . import parser
 from .default import DatasetConfig, EvalConfig, PeftConfig, WandBConfig
 from .policies import PreTrainedConfig
+from .rewards import RewardModelConfig
 
 TRAIN_CONFIG_NAME = "train_config.json"
 
 
+def _migrate_legacy_rabc_fields(config: dict[str, Any]) -> dict[str, Any] | None:
+    """Return migrated payload for legacy RA-BC fields, or None when no migration is needed."""
+    legacy_fields = (
+        "use_rabc",
+        "rabc_progress_path",
+        "rabc_kappa",
+        "rabc_epsilon",
+        "rabc_head_mode",
+    )
+    if not any(key in config for key in legacy_fields):
+        return None
+
+    migrated_config = dict(config)
+    use_rabc = bool(migrated_config.pop("use_rabc", False))
+    rabc_progress_path = migrated_config.pop("rabc_progress_path", None)
+    rabc_kappa = migrated_config.pop("rabc_kappa", None)
+    rabc_epsilon = migrated_config.pop("rabc_epsilon", None)
+    rabc_head_mode = migrated_config.pop("rabc_head_mode", None)
+
+    # New configs may already define sample_weighting explicitly. In that case,
+    # legacy fields are ignored after being stripped from the payload.
+    if migrated_config.get("sample_weighting") is None and use_rabc:
+        sample_weighting: dict[str, Any] = {"type": "rabc"}
+        if rabc_progress_path is not None:
+            sample_weighting["progress_path"] = rabc_progress_path
+        if rabc_kappa is not None:
+            sample_weighting["kappa"] = rabc_kappa
+        if rabc_epsilon is not None:
+            sample_weighting["epsilon"] = rabc_epsilon
+        if rabc_head_mode is not None:
+            sample_weighting["head_mode"] = rabc_head_mode
+        migrated_config["sample_weighting"] = sample_weighting
+
+    return migrated_config
+
+
 @dataclass
 class TrainPipelineConfig(HubMixin):
     dataset: DatasetConfig
     env: envs.EnvConfig | None = None
     policy: PreTrainedConfig | None = None
+    reward_model: RewardModelConfig | None = None
     # Set `dir` to where you would like to save all of the run outputs. If you run another training session
     # with the same value for `dir` its contents will be overwritten unless you set `resume` to true.
     output_dir: Path | None = None
@@ -72,27 +113,44 @@ class TrainPipelineConfig(HubMixin):
     wandb: WandBConfig = field(default_factory=WandBConfig)
     peft: PeftConfig | None = None
 
-    # RA-BC (Reward-Aligned Behavior Cloning) parameters
-    use_rabc: bool = False  # Enable reward-weighted training
-    rabc_progress_path: str | None = None  # Path to precomputed SARM progress parquet file
-    rabc_kappa: float = 0.01  # Hard threshold for high-quality samples
-    rabc_epsilon: float = 1e-6  # Small constant for numerical stability
-    rabc_head_mode: str | None = "sparse"  # For dual-head models: "sparse" or "dense"
+    # Sample weighting configuration (e.g., for RA-BC training)
+    sample_weighting: SampleWeightingConfig | None = None
 
     # Rename map for the observation to override the image and state keys
     rename_map: dict[str, str] = field(default_factory=dict)
     checkpoint_path: Path | None = field(init=False, default=None)
 
+    @property
+    def is_reward_model_training(self) -> bool:
+        """True when the config targets a reward model rather than a policy."""
+        return self.reward_model is not None
+
+    @property
+    def trainable_config(self) -> PreTrainedConfig | RewardModelConfig:
+        """Return whichever config (policy or reward_model) is active."""
+        if self.is_reward_model_training:
+            return self.reward_model  # type: ignore[return-value]
+        return self.policy  # type: ignore[return-value]
+
     def validate(self) -> None:
         # HACK: We parse again the cli args here to get the pretrained paths if there was some.
         policy_path = parser.get_path_arg("policy")
-        if policy_path:
-            # Only load the policy config
-            cli_overrides = parser.get_cli_overrides("policy")
-            self.policy = PreTrainedConfig.from_pretrained(policy_path, cli_overrides=cli_overrides)
+        reward_model_path = parser.get_path_arg("reward_model")
+
+        if reward_model_path:
+            cli_overrides = parser.get_cli_overrides("reward_model")
+            self.reward_model = RewardModelConfig.from_pretrained(
+                reward_model_path, cli_overrides=cli_overrides
+            )
+            self.reward_model.pretrained_path = str(Path(reward_model_path))
+        elif policy_path:
+            yaml_overrides = parser.get_yaml_overrides("policy")
+            cli_overrides = parser.get_cli_overrides("policy") or []
+            self.policy = PreTrainedConfig.from_pretrained(
+                policy_path, cli_overrides=yaml_overrides + cli_overrides
+            )
             self.policy.pretrained_path = Path(policy_path)
         elif self.resume:
-            # The entire train config is already loaded, we just need to get the checkpoint dir
             config_path = parser.parse_arg("config_path")
             if not config_path:
                 raise ValueError(
@@ -108,18 +166,22 @@ class TrainPipelineConfig(HubMixin):
             policy_dir = Path(config_path).parent
             if self.policy is not None:
                 self.policy.pretrained_path = policy_dir
+            if self.reward_model is not None:
+                self.reward_model.pretrained_path = str(policy_dir)
             self.checkpoint_path = policy_dir.parent
 
-        if self.policy is None:
+        if self.policy is None and self.reward_model is None:
             raise ValueError(
-                "Policy is not configured. Please specify a pretrained policy with `--policy.path`."
+                "Neither policy nor reward_model is configured. "
+                "Please specify one with `--policy.path` or `--reward_model.path`."
             )
 
+        active_cfg = self.trainable_config
         if not self.job_name:
             if self.env is None:
-                self.job_name = f"{self.policy.type}"
+                self.job_name = f"{active_cfg.type}"
             else:
-                self.job_name = f"{self.env.type}_{self.policy.type}"
+                self.job_name = f"{self.env.type}_{active_cfg.type}"
 
         if not self.resume and isinstance(self.output_dir, Path) and self.output_dir.is_dir():
             raise FileExistsError(
@@ -137,26 +199,16 @@ class TrainPipelineConfig(HubMixin):
         if not self.use_policy_training_preset and (self.optimizer is None or self.scheduler is None):
             raise ValueError("Optimizer and Scheduler must be set when the policy presets are not used.")
         elif self.use_policy_training_preset and not self.resume:
-            self.optimizer = self.policy.get_optimizer_preset()
-            self.scheduler = self.policy.get_scheduler_preset()
+            self.optimizer = active_cfg.get_optimizer_preset()
+            self.scheduler = active_cfg.get_scheduler_preset()
 
-        if self.policy.push_to_hub and not self.policy.repo_id:
-            raise ValueError(
-                "'policy.repo_id' argument missing. Please specify it to push the model to the hub."
-            )
-
-        if self.use_rabc and not self.rabc_progress_path:
-            # Auto-detect from dataset path
-            repo_id = self.dataset.repo_id
-            if self.dataset.root:
-                self.rabc_progress_path = str(Path(self.dataset.root) / "sarm_progress.parquet")
-            else:
-                self.rabc_progress_path = f"hf://datasets/{repo_id}/sarm_progress.parquet"
+        if hasattr(active_cfg, "push_to_hub") and active_cfg.push_to_hub and not active_cfg.repo_id:
+            raise ValueError("'repo_id' argument missing. Please specify it to push the model to the hub.")
 
     @classmethod
     def __get_path_fields__(cls) -> list[str]:
-        """This enables the parser to load config from the policy using `--policy.path=local/dir`"""
-        return ["policy"]
+        """Keys for draccus pretrained-path loading."""
+        return ["policy", "reward_model"]
 
     def to_dict(self) -> dict[str, Any]:
         return draccus.encode(self)  # type: ignore[no-any-return]  # because of the third-party library draccus uses Any as the return type
@@ -207,12 +259,16 @@ class TrainPipelineConfig(HubMixin):
                 ) from e
 
         cli_args = kwargs.pop("cli_args", [])
+        # Legacy RA-BC migration only applies to framework-saved checkpoints (always JSON).
+        # Hand-written YAML/TOML configs are expected to use the current sample_weighting schema.
+        if config_file is not None and config_file.endswith(".json"):
+            with open(config_file) as f:
+                config = json.load(f)
+            migrated_config = _migrate_legacy_rabc_fields(config)
+            if migrated_config is not None:
+                with tempfile.NamedTemporaryFile("w+", delete=False, suffix=".json") as f:
+                    json.dump(migrated_config, f)
+                    config_file = f.name
+
         with draccus.config_type("json"):
             return draccus.parse(cls, config_file, args=cli_args)
-
-
-@dataclass(kw_only=True)
-class TrainRLServerPipelineConfig(TrainPipelineConfig):
-    # NOTE: In RL, we don't need an offline dataset
-    # TODO: Make `TrainPipelineConfig.dataset` optional
-    dataset: DatasetConfig | None = None  # type: ignore[assignment] # because the parent class has made it's type non-optional

src/lerobot/configs/video.py

@@ -0,0 +1,235 @@
+# Copyright 2026 The HuggingFace Inc. team. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+# Note: We subclass str so that serialization is straightforward
+# https://stackoverflow.com/questions/24481852/serialising-an-enum-member-to-json
+
+"""Video encoder configurations."""
+
+from __future__ import annotations
+
+import logging
+from dataclasses import dataclass, field
+from typing import Any
+
+from lerobot.utils.import_utils import require_package
+
+logger = logging.getLogger(__name__)
+
+# List of hardware encoders to probe for auto-selection. Availability depends on the platform and the chosen video backend.
+# Determines the order of preference for auto-selection when vcodec="auto" is used.
+HW_VIDEO_CODECS = [
+    "h264_videotoolbox",  # macOS
+    "hevc_videotoolbox",  # macOS
+    "h264_nvenc",  # NVIDIA GPU
+    "hevc_nvenc",  # NVIDIA GPU
+    "h264_vaapi",  # Linux Intel/AMD
+    "h264_qsv",  # Intel Quick Sync
+]
+VALID_VIDEO_CODECS: frozenset[str] = frozenset({"h264", "hevc", "libsvtav1", "auto", *HW_VIDEO_CODECS})
+# Aliases for legacy video codec names.
+VIDEO_CODECS_ALIASES: dict[str, str] = {"av1": "libsvtav1"}
+
+
+LIBSVTAV1_DEFAULT_PRESET: int = 12
+
+# Keys persisted under ``features[*]["info"]`` as ``video.<name>`` (from :class:`VideoEncoderConfig`).
+# ``vcodec``` and ``pix_fmt`` are derived from the video stream directly.
+VIDEO_ENCODER_INFO_FIELD_NAMES: frozenset[str] = frozenset(
+    {"g", "crf", "preset", "fast_decode", "extra_options", "video_backend"}
+)
+VIDEO_ENCODER_INFO_KEYS: frozenset[str] = frozenset(
+    f"video.{name}" for name in VIDEO_ENCODER_INFO_FIELD_NAMES
+)
+
+
+@dataclass
+class VideoEncoderConfig:
+    """Video encoder configuration.
+
+    Attributes:
+        vcodec: Video encoder name. ``"auto"`` is resolved during
+            construction (HW encoder if available, else ``libsvtav1``).
+        pix_fmt: Pixel format (e.g. ``"yuv420p"``).
+        g: GOP size (keyframe interval).
+        crf: Quality level — mapped to the native quality parameter of the
+            codec (``crf`` for software, ``qp`` for NVENC/VAAPI,
+            ``q:v`` for VideoToolbox, ``global_quality`` for QSV).
+        preset: Speed/quality preset. Accepted type is per-codec.
+        fast_decode: Fast-decode tuning. For ``libsvtav1`` this is a level (0-2)
+            embedded in ``svtav1-params``. For ``h264`` and ``hevc`` non-zero values
+            set ``tune=fastdecode``. Ignored for other codecs.
+        video_backend: Python to be used for encoding. Only ``"pyav"``
+            is currently supported.
+        extra_options: Free-form dictionary of additional video encoder options
+            (e.g. ``{"tune": "film", "profile:v": "high", "bf": 2}``).
+    """
+
+    vcodec: str = "libsvtav1"  # TODO(CarolinePascal): rename to codec ?
+    pix_fmt: str = "yuv420p"
+    g: int | None = 2
+    crf: int | float | None = 30
+    preset: int | str | None = None
+    fast_decode: int = 0
+    # TODO(CarolinePascal): add torchcodec support + find a way to unify the
+    # two backends (encoding and decoding).
+    video_backend: str = "pyav"
+    extra_options: dict[str, Any] = field(default_factory=dict)
+
+    def __post_init__(self) -> None:
+        self.resolve_vcodec()
+        # Empty-constructor ergonomics: ``VideoEncoderConfig()`` must "just work".
+        if self.preset is None and self.vcodec == "libsvtav1":
+            self.preset = LIBSVTAV1_DEFAULT_PRESET
+        self.validate()
+
+    @classmethod
+    def from_video_info(cls, video_info: dict | None) -> VideoEncoderConfig:
+        """Reconstruct a :class:`VideoEncoderConfig` from a video feature's ``info`` block.
+        Missing or ``None`` values fall back to the class defaults.
+        """
+        video_info = video_info or {}
+        kwargs: dict[str, Any] = {}
+
+        for src_key, dst_field in (("video.codec", "vcodec"), ("video.pix_fmt", "pix_fmt")):
+            value = video_info.get(src_key)
+            if value is not None:
+                kwargs[dst_field] = value
+
+        for field_name in VIDEO_ENCODER_INFO_FIELD_NAMES:
+            value = video_info.get(f"video.{field_name}")
+            if value is None:
+                continue
+            # Persisted as ``{}`` after merges with disagreeing sources — treat as default.
+            if field_name == "extra_options" and not value:
+                continue
+            kwargs[field_name] = value
+
+        return cls(**kwargs)
+
+    def detect_available_encoders(self, encoders: list[str] | str) -> list[str]:
+        """Return the subset of available encoders based on the specified video backend.
+
+        Args:
+            encoders: List of encoder names to detect. If a string, it is converted to a list.
+        Returns:
+            List of available encoder names. If the video backend is not "pyav", returns an empty list.
+        """
+        if self.video_backend == "pyav":
+            require_package("av", extra="dataset")
+            from lerobot.datasets import detect_available_encoders_pyav
+
+            return detect_available_encoders_pyav(encoders)
+        return []
+
+    def validate(self) -> None:
+        """Validate the video encoder configuration."""
+        if self.video_backend == "pyav":
+            require_package("av", extra="dataset")
+            from lerobot.datasets import check_video_encoder_parameters_pyav
+
+            check_video_encoder_parameters_pyav(self.vcodec, self.pix_fmt, self.get_codec_options())
+
+    def resolve_vcodec(self) -> None:
+        """Check ``vcodec`` and, when it is ``"auto"``, pick a concrete encoder.
+
+        For ``"auto"``, the first hardware encoder in the preference list that is available is chosen; if none are available, ``libsvtav1`` is used. If the
+        resolved codec (explicit or after auto-selection) is not available, raises ``ValueError``.
+
+        Stream-derived canonical codec names listed in :data:`VIDEO_CODECS_ALIASES` are
+        rewritten to their corresponding encoder name (e.g. ``"av1"`` → ``"libsvtav1"``).
+        """
+        self.vcodec = VIDEO_CODECS_ALIASES.get(self.vcodec, self.vcodec)
+        if self.vcodec not in VALID_VIDEO_CODECS:
+            raise ValueError(f"Invalid vcodec '{self.vcodec}'. Must be one of: {sorted(VALID_VIDEO_CODECS)}")
+        if self.vcodec == "auto":
+            available = self.detect_available_encoders(HW_VIDEO_CODECS)
+            for encoder in HW_VIDEO_CODECS:
+                if encoder in available:
+                    logger.info(f"Auto-selected video codec: {encoder}")
+                    self.vcodec = encoder
+                    return
+            logger.warning("No hardware encoder available, falling back to software encoder 'libsvtav1'")
+            self.vcodec = "libsvtav1"
+
+        if self.detect_available_encoders(self.vcodec):
+            logger.info(f"Using video codec: {self.vcodec}")
+            return
+        raise ValueError(f"Unsupported video codec: {self.vcodec} with video backend {self.video_backend}")
+
+    def get_codec_options(
+        self, encoder_threads: int | None = None, as_strings: bool = False
+    ) -> dict[str, Any]:
+        """Translate the tuning fields to codec-specific options.
+
+        ``VideoEncoderConfig.extra_options`` are merged last but never override a structured field.
+
+        Args:
+            encoder_threads: Number of encoder threads set globally for all VideoEncoderConfigs.
+                For libsvtav1, this is mapped to ``lp`` via ``svtav1-params``.
+                For h264/hevc, this is mapped to ``threads``.
+                Hardware encoders ignore this parameter.
+            as_strings: If ``True``, casts values to strings.
+        """
+        opts: dict[str, Any] = {}
+
+        def set_if(key: str, value: Any) -> None:
+            if value is not None:
+                opts[key] = value if not as_strings else str(value)
+
+        # GOP size is not a codec-specific option, so it is always set.
+        set_if("g", self.g)
+
+        if self.vcodec == "libsvtav1":
+            set_if("crf", self.crf)
+            set_if("preset", self.preset)
+            svtav1_parts: list[str] = []
+            if self.fast_decode is not None:
+                svtav1_parts.append(f"fast-decode={max(0, min(2, self.fast_decode))}")
+            if encoder_threads is not None:
+                svtav1_parts.append(f"lp={encoder_threads}")
+            if svtav1_parts:
+                opts["svtav1-params"] = ":".join(svtav1_parts)
+        elif self.vcodec in ("h264", "hevc"):
+            set_if("crf", self.crf)
+            set_if("preset", self.preset)
+            if self.fast_decode:
+                opts["tune"] = "fastdecode"
+            set_if("threads", encoder_threads)
+        elif self.vcodec in ("h264_videotoolbox", "hevc_videotoolbox"):
+            if self.crf is not None:
+                opts["q:v"] = max(1, min(100, 100 - self.crf * 2))
+        elif self.vcodec in ("h264_nvenc", "hevc_nvenc"):
+            opts["rc"] = 0
+            set_if("qp", self.crf)
+            set_if("preset", self.preset)
+        elif self.vcodec == "h264_vaapi":
+            set_if("qp", self.crf)
+        elif self.vcodec == "h264_qsv":
+            set_if("global_quality", self.crf)
+            set_if("preset", self.preset)
+        else:
+            set_if("crf", self.crf)
+            set_if("preset", self.preset)
+
+        # Extra options are merged last but never override structured fields (values are kept as given).
+        for k, v in self.extra_options.items():
+            if k not in opts:
+                set_if(k, v)
+
+        return opts
+
+
+def camera_encoder_defaults() -> VideoEncoderConfig:
+    """Return a :class:`VideoEncoderConfig` with RGB-camera defaults."""
+    return VideoEncoderConfig()

src/lerobot/datasets/__init__.py

@@ -31,15 +31,25 @@ from .dataset_tools import (
     modify_features,
     modify_tasks,
     recompute_stats,
+    reencode_dataset,
     remove_feature,
     split_dataset,
 )
 from .factory import make_dataset, resolve_delta_timestamps
 from .image_writer import safe_stop_image_writer
 from .io_utils import load_episodes, write_stats
+from .language import (
+    EVENT_ONLY_STYLES,
+    LANGUAGE_EVENTS,
+    LANGUAGE_PERSISTENT,
+    PERSISTENT_STYLES,
+    STYLE_REGISTRY,
+    column_for_style,
+)
 from .lerobot_dataset import LeRobotDataset
 from .multi_dataset import MultiLeRobotDataset
 from .pipeline_features import aggregate_pipeline_dataset_features, create_initial_features
+from .pyav_utils import check_video_encoder_parameters_pyav, detect_available_encoders_pyav
 from .sampler import EpisodeAwareSampler
 from .streaming_dataset import StreamingLeRobotDataset
 from .utils import DEFAULT_EPISODES_PATH, create_lerobot_dataset_card
@@ -53,12 +63,19 @@ __all__ = [
     "CODEBASE_VERSION",
     "DEFAULT_EPISODES_PATH",
     "DEFAULT_QUANTILES",
+    "EVENT_ONLY_STYLES",
     "EpisodeAwareSampler",
+    "LANGUAGE_EVENTS",
+    "LANGUAGE_PERSISTENT",
     "LeRobotDataset",
     "LeRobotDatasetMetadata",
     "MultiLeRobotDataset",
+    "PERSISTENT_STYLES",
+    "STYLE_REGISTRY",
     "StreamingLeRobotDataset",
     "VideoEncodingManager",
+    "check_video_encoder_parameters_pyav",
+    "detect_available_encoders_pyav",
     "add_features",
     "aggregate_datasets",
     "aggregate_pipeline_dataset_features",
@@ -66,6 +83,7 @@ __all__ = [
     "convert_image_to_video_dataset",
     "create_initial_features",
     "create_lerobot_dataset_card",
+    "column_for_style",
     "delete_episodes",
     "get_feature_stats",
     "load_episodes",
@@ -74,6 +92,7 @@ __all__ = [
     "modify_features",
     "modify_tasks",
     "recompute_stats",
+    "reencode_dataset",
     "remove_feature",
     "resolve_delta_timestamps",
     "safe_stop_image_writer",

src/lerobot/datasets/aggregate.py

@@ -15,6 +15,7 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
+import copy
 import logging
 import shutil
 from pathlib import Path
@@ -23,9 +24,11 @@ import datasets
 import pandas as pd
 import tqdm
 
+from lerobot.configs import VIDEO_ENCODER_INFO_KEYS
+
 from .compute_stats import aggregate_stats
 from .dataset_metadata import LeRobotDatasetMetadata
-from .feature_utils import get_hf_features_from_features
+from .feature_utils import features_equal_for_merge, get_hf_features_from_features
 from .io_utils import (
     get_file_size_in_mb,
     get_parquet_file_size_in_mb,
@@ -46,11 +49,54 @@ from .utils import (
 from .video_utils import concatenate_video_files, get_video_duration_in_s
 
 
+def merge_video_feature_info_for_aggregate(all_metadata: list[LeRobotDatasetMetadata]) -> dict[str, dict]:
+    """Create a merged video feature info dictionary for aggregation. The video encoder info is merged field-by-field: each key is kept only when every source agrees; otherwise that key is set to ``null`` (or ``{}`` for ``video.extra_options``) and a warning is logged.
+
+    Args:
+        all_metadata: List of LeRobotDatasetMetadata objects to merge.
+
+    Returns:
+        dict: A dictionary of merged video feature info.
+    """
+    merged_info = copy.deepcopy(all_metadata[0].features)
+    video_keys = [k for k in merged_info if merged_info[k].get("dtype") == "video"]
+
+    for vk in video_keys:
+        video_infos = [m.features.get(vk, {}).get("info") or {} for m in all_metadata]
+        base_video_info = video_infos[0]
+
+        merged_encoder_info: dict = {}
+        fallback_keys: list[str] = []
+        for info_key in VIDEO_ENCODER_INFO_KEYS:
+            values = [info.get(info_key, None) for info in video_infos]
+            first_value = values[0]
+            all_match = all(v == first_value for v in values[1:])
+
+            if all_match:
+                merged_encoder_info[info_key] = first_value
+            else:
+                fallback_keys.append(info_key)
+                merged_encoder_info[info_key] = {} if info_key == "video.extra_options" else None
+
+        if fallback_keys:
+            logging.warning(
+                f"Merging heterogeneous or incomplete video encoder metadata for feature {vk}. "
+                f"Setting these keys to null: {fallback_keys}.",
+            )
+
+        merged_info[vk]["info"] = {**base_video_info, **merged_encoder_info}
+        # TODO(CarolinePascal): make this variable once we have support for other video backends.
+        merged_info[vk]["info"]["video.video_backend"] = "pyav"
+
+    return merged_info
+
+
 def validate_all_metadata(all_metadata: list[LeRobotDatasetMetadata]):
     """Validates that all dataset metadata have consistent properties.
 
     Ensures all datasets have the same fps, robot_type, and features to guarantee
     compatibility when aggregating them into a single dataset.
+    Video encoder info is not considered for validation but is merged during aggregation in ``merge_video_feature_info_for_aggregate``.
 
     Args:
         all_metadata: List of LeRobotDatasetMetadata objects to validate.
@@ -74,7 +120,7 @@ def validate_all_metadata(all_metadata: list[LeRobotDatasetMetadata]):
             raise ValueError(
                 f"Same robot_type is expected, but got robot_type={meta.robot_type} instead of {robot_type}."
             )
-        if features != meta.features:
+        if not features_equal_for_merge(features, meta.features):
             raise ValueError(
                 f"Same features is expected, but got features={meta.features} instead of {features}."
             )
@@ -97,8 +143,8 @@ def update_data_df(df, src_meta, dst_meta):
         pd.DataFrame: Updated DataFrame with adjusted indices.
     """
 
-    df["episode_index"] = df["episode_index"] + dst_meta.info["total_episodes"]
-    df["index"] = df["index"] + dst_meta.info["total_frames"]
+    df["episode_index"] = df["episode_index"] + dst_meta.info.total_episodes
+    df["index"] = df["index"] + dst_meta.info.total_frames
 
     src_task_names = src_meta.tasks.index.take(df["task_index"].to_numpy())
     df["task_index"] = dst_meta.tasks.loc[src_task_names, "task_index"].to_numpy()
@@ -225,9 +271,9 @@ def update_meta_data(
         # Clean up temporary columns
         df = df.drop(columns=["_orig_chunk", "_orig_file"])
 
-    df["dataset_from_index"] = df["dataset_from_index"] + dst_meta.info["total_frames"]
-    df["dataset_to_index"] = df["dataset_to_index"] + dst_meta.info["total_frames"]
-    df["episode_index"] = df["episode_index"] + dst_meta.info["total_episodes"]
+    df["dataset_from_index"] = df["dataset_from_index"] + dst_meta.info.total_frames
+    df["dataset_to_index"] = df["dataset_to_index"] + dst_meta.info.total_frames
+    df["episode_index"] = df["episode_index"] + dst_meta.info.total_episodes
 
     return df
 
@@ -237,8 +283,8 @@ def aggregate_datasets(
     aggr_repo_id: str,
     roots: list[Path] | None = None,
     aggr_root: Path | None = None,
-    data_files_size_in_mb: float | None = None,
-    video_files_size_in_mb: float | None = None,
+    data_files_size_in_mb: int | None = None,
+    video_files_size_in_mb: int | None = None,
     chunk_size: int | None = None,
 ):
     """Aggregates multiple LeRobot datasets into a single unified dataset.
@@ -274,7 +320,8 @@ def aggregate_datasets(
             LeRobotDatasetMetadata(repo_id, root=root) for repo_id, root in zip(repo_ids, roots, strict=False)
         ]
     )
-    fps, robot_type, features = validate_all_metadata(all_metadata)
+    fps, robot_type, _ = validate_all_metadata(all_metadata)
+    features = merge_video_feature_info_for_aggregate(all_metadata)
     video_keys = [key for key in features if features[key]["dtype"] == "video"]
 
     dst_meta = LeRobotDatasetMetadata.create(
@@ -313,8 +360,8 @@ def aggregate_datasets(
         # to avoid interference between different source datasets
         data_idx.pop("src_to_dst", None)
 
-        dst_meta.info["total_episodes"] += src_meta.total_episodes
-        dst_meta.info["total_frames"] += src_meta.total_frames
+        dst_meta.info.total_episodes += src_meta.total_episodes
+        dst_meta.info.total_frames += src_meta.total_frames
 
     finalize_aggregation(dst_meta, all_metadata)
     logging.info("Aggregation complete.")
@@ -332,7 +379,6 @@ def aggregate_videos(src_meta, dst_meta, videos_idx, video_files_size_in_mb, chu
         videos_idx: Dictionary tracking video chunk and file indices.
         video_files_size_in_mb: Maximum size for video files in MB (defaults to DEFAULT_VIDEO_FILE_SIZE_IN_MB)
         chunk_size: Maximum number of files per chunk (defaults to DEFAULT_CHUNK_SIZE)
-
     Returns:
         dict: Updated videos_idx with current chunk and file indices.
     """
@@ -414,9 +460,11 @@ def aggregate_videos(src_meta, dst_meta, videos_idx, video_files_size_in_mb, chu
                 current_dst_duration = dst_file_durations.get(dst_key, 0)
                 videos_idx[key]["src_to_offset"][(src_chunk_idx, src_file_idx)] = current_dst_duration
                 videos_idx[key]["src_to_dst"][(src_chunk_idx, src_file_idx)] = dst_key
+                # TODO(CarolinePascal): Move the check before the loop to avoid failing in the middle + add possibility to re-encode the video if the check fails
                 concatenate_video_files(
                     [dst_path, src_path],
                     dst_path,
+                    compatibility_check=True,
                 )
                 # Update duration of this destination file
                 dst_file_durations[dst_key] = current_dst_duration + src_duration
@@ -640,14 +688,10 @@ def finalize_aggregation(aggr_meta, all_metadata):
     write_tasks(aggr_meta.tasks, aggr_meta.root)
 
     logging.info("write info")
-    aggr_meta.info.update(
-        {
-            "total_tasks": len(aggr_meta.tasks),
-            "total_episodes": sum(m.total_episodes for m in all_metadata),
-            "total_frames": sum(m.total_frames for m in all_metadata),
-            "splits": {"train": f"0:{sum(m.total_episodes for m in all_metadata)}"},
-        }
-    )
+    aggr_meta.info.total_tasks = len(aggr_meta.tasks)
+    aggr_meta.info.total_episodes = sum(m.total_episodes for m in all_metadata)
+    aggr_meta.info.total_frames = sum(m.total_frames for m in all_metadata)
+    aggr_meta.info.splits = {"train": f"0:{sum(m.total_episodes for m in all_metadata)}"}
     write_info(aggr_meta.info, aggr_meta.root)
 
     logging.info("write stats")

src/lerobot/datasets/compute_stats.py

@@ -512,7 +512,7 @@ def compute_episode_stats(
 
     ep_stats = {}
     for key, data in episode_data.items():
-        if features[key]["dtype"] == "string":
+        if features[key]["dtype"] in {"string", "language"}:
             continue
 
         if features[key]["dtype"] in ["image", "video"]:

src/lerobot/datasets/dataset_metadata.py

@@ -14,6 +14,7 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 import contextlib
+from collections.abc import Callable
 from pathlib import Path
 
 import numpy as np
@@ -23,6 +24,7 @@ import pyarrow as pa
 import pyarrow.parquet as pq
 from huggingface_hub import snapshot_download
 
+from lerobot.configs import VideoEncoderConfig
 from lerobot.utils.constants import DEFAULT_FEATURES, HF_LEROBOT_HOME, HF_LEROBOT_HUB_CACHE
 from lerobot.utils.feature_utils import _validate_feature_names
 from lerobot.utils.utils import flatten_dict
@@ -34,16 +36,14 @@ from .io_utils import (
     load_episodes,
     load_info,
     load_stats,
-    load_subtasks,
     load_tasks,
     write_info,
-    write_json,
     write_stats,
     write_tasks,
 )
+from .language import DEFAULT_TOOLS, LANGUAGE_COLUMNS
 from .utils import (
     DEFAULT_EPISODES_PATH,
-    INFO_PATH,
     check_version_compatibility,
     get_safe_version,
     has_legacy_hub_download_metadata,
@@ -177,7 +177,6 @@ class LeRobotDatasetMetadata:
         self.info = load_info(self.root)
         check_version_compatibility(self.repo_id, self._version, CODEBASE_VERSION)
         self.tasks = load_tasks(self.root)
-        self.subtasks = load_subtasks(self.root)
         self.episodes = load_episodes(self.root)
         self.stats = load_stats(self.root)
 
@@ -191,6 +190,29 @@ class LeRobotDatasetMetadata:
         if self.episodes is None:
             self._load_metadata()
 
+    def filter_episodes(
+        self,
+        predicate: Callable[[dict], bool],
+        candidates: list[int] | None = None,
+    ) -> list[int]:
+        """Filter episodes whose metadata satisfies a given predicate.
+
+        Args:
+            predicate: Predicate over per-episode metadata rows used to select episodes.
+            candidates: Optional list of episode indices to restrict evaluation to.
+
+        Returns:
+            List of sorted episode indices that satisfy the predicate.
+        """
+        self.ensure_readable()
+        if candidates is not None:
+            candidate_set = set(candidates)
+            combined = lambda ep: ep["episode_index"] in candidate_set and predicate(ep)  # noqa: E731
+        else:
+            combined = predicate
+        filtered = self.episodes.filter(combined, keep_in_memory=True, load_from_cache_file=False)
+        return sorted(int(idx) for idx in filtered["episode_index"])
+
     def _pull_from_repo(
         self,
         allow_patterns: list[str] | str | None = None,
@@ -228,7 +250,7 @@ class LeRobotDatasetMetadata:
     @property
     def _version(self) -> packaging.version.Version:
         """Codebase version used to create this dataset."""
-        return packaging.version.parse(self.info["codebase_version"])
+        return packaging.version.parse(self.info.codebase_version)
 
     def get_data_file_path(self, ep_index: int) -> Path:
         """Return the relative parquet file path for the given episode index.
@@ -283,27 +305,27 @@ class LeRobotDatasetMetadata:
     @property
     def data_path(self) -> str:
         """Formattable string for the parquet files."""
-        return self.info["data_path"]
+        return self.info.data_path
 
     @property
     def video_path(self) -> str | None:
         """Formattable string for the video files."""
-        return self.info["video_path"]
+        return self.info.video_path
 
     @property
     def robot_type(self) -> str | None:
         """Robot type used in recording this dataset."""
-        return self.info["robot_type"]
+        return self.info.robot_type
 
     @property
     def fps(self) -> int:
         """Frames per second used during data collection."""
-        return self.info["fps"]
+        return self.info.fps
 
     @property
     def features(self) -> dict[str, dict]:
         """All features contained in the dataset."""
-        return self.info["features"]
+        return self.info.features
 
     @property
     def image_keys(self) -> list[str]:
@@ -320,6 +342,49 @@ class LeRobotDatasetMetadata:
         """Keys to access visual modalities (regardless of their storage method)."""
         return [key for key, ft in self.features.items() if ft["dtype"] in ["video", "image"]]
 
+    @property
+    def has_language_columns(self) -> bool:
+        """Return ``True`` if the dataset declares any language column.
+
+        Used to gate language-aware code paths (collate, render step) so
+        unannotated datasets keep PyTorch's default collate behavior.
+        """
+        return any(col in self.features for col in LANGUAGE_COLUMNS)
+
+    @property
+    def tools(self) -> list[dict]:
+        """OpenAI-style tool schemas declared by this dataset.
+
+        Read from ``meta/info.json["tools"]``. Returns a copy, so callers
+        can mutate the result safely. Falls back to
+        :data:`lerobot.datasets.language.DEFAULT_TOOLS` (the canonical
+        ``say`` schema) when the dataset doesn't declare any — that way
+        unannotated datasets and chat-template consumers
+        (``apply_chat_template(messages, tools=meta.tools)``) keep
+        working out of the box.
+
+        Implementations live under :mod:`lerobot.tools` (one file per
+        tool); see ``docs/source/tools.mdx`` for the authoring guide.
+        """
+        declared = self.info.tools
+        if declared:
+            return [dict(t) for t in declared]
+        return [dict(t) for t in DEFAULT_TOOLS]
+
+    @tools.setter
+    def tools(self, value: list[dict] | None) -> None:
+        """Persist a tool catalog to ``meta/info.json`` and reload metadata.
+
+        Writes ``value`` into the on-disk ``info.json`` (or clears the
+        ``tools`` key when ``value`` is ``None`` or empty), then reloads
+        ``self.info`` so the in-memory metadata matches what's on disk.
+        Saves callers from hand-editing ``info.json`` and re-instantiating
+        the metadata object.
+        """
+        self.info.tools = [dict(t) for t in value] if value else None
+        write_info(self.info, self.root)
+        self.info = load_info(self.root)
+
     @property
     def names(self) -> dict[str, list | dict]:
         """Names of the various dimensions of vector modalities."""
@@ -333,32 +398,32 @@ class LeRobotDatasetMetadata:
     @property
     def total_episodes(self) -> int:
         """Total number of episodes available."""
-        return self.info["total_episodes"]
+        return self.info.total_episodes
 
     @property
     def total_frames(self) -> int:
         """Total number of frames saved in this dataset."""
-        return self.info["total_frames"]
+        return self.info.total_frames
 
     @property
     def total_tasks(self) -> int:
         """Total number of different tasks performed in this dataset."""
-        return self.info["total_tasks"]
+        return self.info.total_tasks
 
     @property
     def chunks_size(self) -> int:
         """Max number of files per chunk."""
-        return self.info["chunks_size"]
+        return self.info.chunks_size
 
     @property
     def data_files_size_in_mb(self) -> int:
         """Max size of data file in mega bytes."""
-        return self.info["data_files_size_in_mb"]
+        return self.info.data_files_size_in_mb
 
     @property
     def video_files_size_in_mb(self) -> int:
         """Max size of video file in mega bytes."""
-        return self.info["video_files_size_in_mb"]
+        return self.info.video_files_size_in_mb
 
     def get_task_index(self, task: str) -> int | None:
         """
@@ -502,20 +567,33 @@ class LeRobotDatasetMetadata:
         self._save_episode_metadata(episode_dict)
 
         # Update info
-        self.info["total_episodes"] += 1
-        self.info["total_frames"] += episode_length
-        self.info["total_tasks"] = len(self.tasks)
-        self.info["splits"] = {"train": f"0:{self.info['total_episodes']}"}
+        self.info.total_episodes += 1
+        self.info.total_frames += episode_length
+        self.info.total_tasks = len(self.tasks)
+        self.info.splits = {"train": f"0:{self.info.total_episodes}"}
 
         write_info(self.info, self.root)
 
         self.stats = aggregate_stats([self.stats, episode_stats]) if self.stats is not None else episode_stats
         write_stats(self.stats, self.root)
 
-    def update_video_info(self, video_key: str | None = None) -> None:
-        """
+    def update_video_info(
+        self,
+        video_key: str | None = None,
+        camera_encoder: VideoEncoderConfig | None = None,
+    ) -> None:
+        """Populate per-feature video info in ``info.json``.
+
         Warning: this function writes info from first episode videos, implicitly assuming that all videos have
         been encoded the same way. Also, this means it assumes the first episode exists.
+
+        Args:
+            video_key: If provided, only update this video key. Otherwise update
+                all video keys in the dataset.
+            camera_encoder: Encoder configuration used to produce the
+                videos. When provided, its fields are recorded as
+                ``video.<field>`` entries alongside the stream-derived
+                ``video.*`` entries (see :func:`get_video_info`).
         """
         if video_key is not None and video_key not in self.video_keys:
             raise ValueError(f"Video key {video_key} not found in dataset")
@@ -524,7 +602,7 @@ class LeRobotDatasetMetadata:
         for key in video_keys:
             if not self.features[key].get("info", None):
                 video_path = self.root / self.video_path.format(video_key=key, chunk_index=0, file_index=0)
-                self.info["features"][key]["info"] = get_video_info(video_path)
+                self.info.features[key]["info"] = get_video_info(video_path, camera_encoder=camera_encoder)
 
     def update_chunk_settings(
         self,
@@ -546,17 +624,17 @@ class LeRobotDatasetMetadata:
         if chunks_size is not None:
             if chunks_size <= 0:
                 raise ValueError(f"chunks_size must be positive, got {chunks_size}")
-            self.info["chunks_size"] = chunks_size
+            self.info.chunks_size = chunks_size
 
         if data_files_size_in_mb is not None:
             if data_files_size_in_mb <= 0:
                 raise ValueError(f"data_files_size_in_mb must be positive, got {data_files_size_in_mb}")
-            self.info["data_files_size_in_mb"] = data_files_size_in_mb
+            self.info.data_files_size_in_mb = data_files_size_in_mb
 
         if video_files_size_in_mb is not None:
             if video_files_size_in_mb <= 0:
                 raise ValueError(f"video_files_size_in_mb must be positive, got {video_files_size_in_mb}")
-            self.info["video_files_size_in_mb"] = video_files_size_in_mb
+            self.info.video_files_size_in_mb = video_files_size_in_mb
 
         # Update the info file on disk
         write_info(self.info, self.root)
@@ -635,7 +713,6 @@ class LeRobotDatasetMetadata:
         _validate_feature_names(features)
 
         obj.tasks = None
-        obj.subtasks = None
         obj.episodes = None
         obj.stats = None
         obj.info = create_empty_dataset_info(
@@ -653,7 +730,7 @@ class LeRobotDatasetMetadata:
                 f"Features contain video keys {obj.video_keys}, but 'use_videos' is set to False. "
                 "Either remove video features from the features dict, or set 'use_videos=True'."
             )
-        write_json(obj.info, obj.root / INFO_PATH)
+        write_info(obj.info, obj.root)
         obj.revision = None
         obj._pq_writer = None
         obj.latest_episode = None

src/lerobot/datasets/dataset_reader.py

@@ -295,9 +295,4 @@ class DatasetReader:
         task_idx = item["task_index"].item()
         item["task"] = self._meta.tasks.iloc[task_idx].name
 
-        # add subtask information if available
-        if "subtask_index" in self._meta.features and self._meta.subtasks is not None:
-            subtask_idx = item["subtask_index"].item()
-            item["subtask"] = self._meta.subtasks.iloc[subtask_idx].name
-
         return item

src/lerobot/datasets/dataset_tools.py

@@ -26,7 +26,7 @@ This module provides utilities for:
 import logging
 import shutil
 from collections.abc import Callable
-from concurrent.futures import ThreadPoolExecutor, as_completed
+from concurrent.futures import ProcessPoolExecutor, ThreadPoolExecutor, as_completed
 from pathlib import Path
 
 import datasets
@@ -36,6 +36,7 @@ import pyarrow.parquet as pq
 import torch
 from tqdm import tqdm
 
+from lerobot.configs import VideoEncoderConfig, camera_encoder_defaults
 from lerobot.utils.constants import ACTION, HF_LEROBOT_HOME, OBS_IMAGE, OBS_STATE
 from lerobot.utils.utils import flatten_dict
 
@@ -60,9 +61,14 @@ from .utils import (
     DEFAULT_DATA_FILE_SIZE_IN_MB,
     DEFAULT_DATA_PATH,
     DEFAULT_EPISODES_PATH,
+    VIDEO_DIR,
     update_chunk_file_indices,
 )
-from .video_utils import encode_video_frames, get_video_info
+from .video_utils import (
+    encode_video_frames,
+    get_video_info,
+    reencode_video,
+)
 
 
 def _load_episode_with_stats(src_dataset: LeRobotDataset, episode_idx: int) -> dict:
@@ -95,6 +101,11 @@ def delete_episodes(
 ) -> LeRobotDataset:
     """Delete episodes from a LeRobotDataset and create a new dataset.
 
+    Video segments that need re-encoding (because the source file mixes kept and
+    deleted episodes) are re-encoded with the source dataset's existing encoder
+    settings — read back from ``meta/info.json`` — so the output dataset stays
+    consistent with its own metadata.
+
     Args:
         dataset: The source LeRobotDataset.
         episode_indices: List of episode indices to delete.
@@ -157,6 +168,11 @@ def split_dataset(
 ) -> dict[str, LeRobotDataset]:
     """Split a LeRobotDataset into multiple smaller datasets.
 
+    Video segments that need re-encoding (because the source file mixes episodes
+    that fall into different splits) are re-encoded with the source dataset's
+    existing encoder settings — read back from ``meta/info.json`` — so each
+    output split stays consistent with its own metadata.
+
     Args:
         dataset: The source LeRobotDataset to split.
         splits: Either a dict mapping split names to episode indices, or a dict mapping
@@ -578,8 +594,7 @@ def _keep_episodes_from_video_with_av(
     output_path: Path,
     episodes_to_keep: list[tuple[int, int]],
     fps: float,
-    vcodec: str = "libsvtav1",
-    pix_fmt: str = "yuv420p",
+    camera_encoder: VideoEncoderConfig,
 ) -> None:
     """Keep only specified episodes from a video file using PyAV.
 
@@ -593,8 +608,7 @@ def _keep_episodes_from_video_with_av(
             Ranges are half-open intervals: [start_frame, end_frame), where start_frame
             is inclusive and end_frame is exclusive.
         fps: Frame rate of the video.
-        vcodec: Video codec to use for encoding.
-        pix_fmt: Pixel format for output video.
+        camera_encoder: Video encoder settings used to re-encode the kept frames.
     """
     from fractions import Fraction
 
@@ -619,12 +633,13 @@ def _keep_episodes_from_video_with_av(
 
     # Convert fps to Fraction for PyAV compatibility.
     fps_fraction = Fraction(fps).limit_denominator(1000)
-    v_out = out.add_stream(vcodec, rate=fps_fraction)
+    codec_options = camera_encoder.get_codec_options(as_strings=True)
+    v_out = out.add_stream(camera_encoder.vcodec, rate=fps_fraction, options=codec_options)
 
     # PyAV type stubs don't distinguish video streams from audio/subtitle streams.
     v_out.width = v_in.codec_context.width
     v_out.height = v_in.codec_context.height
-    v_out.pix_fmt = pix_fmt
+    v_out.pix_fmt = camera_encoder.pix_fmt
 
     # Set time_base to match the frame rate for proper timestamp handling.
     v_out.time_base = Fraction(1, int(fps))
@@ -687,14 +702,14 @@ def _copy_and_reindex_videos(
     src_dataset: LeRobotDataset,
     dst_meta: LeRobotDatasetMetadata,
     episode_mapping: dict[int, int],
-    vcodec: str = "libsvtav1",
-    pix_fmt: str = "yuv420p",
 ) -> dict[int, dict]:
     """Copy and filter video files, only re-encoding files with deleted episodes.
 
     For video files that only contain kept episodes, we copy them directly.
     For files with mixed kept/deleted episodes, we use PyAV filters to efficiently
-    re-encode only the desired segments.
+    re-encode only the desired segments. The encoder used for re-encoding is
+    derived per video key from the source dataset's ``meta/info.json`` so the
+    destination metadata keeps describing the videos accurately.
 
     Args:
         src_dataset: Source dataset to copy from
@@ -711,6 +726,9 @@ def _copy_and_reindex_videos(
 
     for video_key in src_dataset.meta.video_keys:
         logging.info(f"Processing videos for {video_key}")
+        camera_encoder = VideoEncoderConfig.from_video_info(
+            src_dataset.meta.info.features.get(video_key, {}).get("info")
+        )
 
         if dst_meta.video_path is None:
             raise ValueError("Destination metadata has no video_path defined")
@@ -792,8 +810,7 @@ def _copy_and_reindex_videos(
                     dst_video_path,
                     episodes_to_keep_ranges,
                     src_dataset.meta.fps,
-                    vcodec,
-                    pix_fmt,
+                    camera_encoder,
                 )
 
                 cumulative_ts = 0.0
@@ -897,14 +914,10 @@ def _copy_and_reindex_episodes_metadata(
 
     dst_meta.finalize()
 
-    dst_meta.info.update(
-        {
-            "total_episodes": len(episode_mapping),
-            "total_frames": total_frames,
-            "total_tasks": len(dst_meta.tasks) if dst_meta.tasks is not None else 0,
-            "splits": {"train": f"0:{len(episode_mapping)}"},
-        }
-    )
+    dst_meta.info.total_episodes = len(episode_mapping)
+    dst_meta.info.total_frames = total_frames
+    dst_meta.info.total_tasks = len(dst_meta.tasks) if dst_meta.tasks is not None else 0
+    dst_meta.info.splits = {"train": f"0:{len(episode_mapping)}"}
     write_info(dst_meta.info, dst_meta.root)
 
     if not all_stats:
@@ -1069,21 +1082,20 @@ def _copy_episodes_metadata_and_stats(
     if episodes_dir.exists():
         shutil.copytree(episodes_dir, dst_episodes_dir, dirs_exist_ok=True)
 
-    dst_meta.info.update(
-        {
-            "total_episodes": src_dataset.meta.total_episodes,
-            "total_frames": src_dataset.meta.total_frames,
-            "total_tasks": src_dataset.meta.total_tasks,
-            "splits": src_dataset.meta.info.get("splits", {"train": f"0:{src_dataset.meta.total_episodes}"}),
-        }
+    dst_meta.info.total_episodes = src_dataset.meta.total_episodes
+    dst_meta.info.total_frames = src_dataset.meta.total_frames
+    dst_meta.info.total_tasks = src_dataset.meta.total_tasks
+    # Preserve original splits if available, otherwise create default
+    dst_meta.info.splits = (
+        src_dataset.meta.info.splits
+        if src_dataset.meta.info.splits
+        else {"train": f"0:{src_dataset.meta.total_episodes}"}
     )
 
     if dst_meta.video_keys and src_dataset.meta.video_keys:
         for key in dst_meta.video_keys:
             if key in src_dataset.meta.features:
-                dst_meta.info["features"][key]["info"] = src_dataset.meta.info["features"][key].get(
-                    "info", {}
-                )
+                dst_meta.info.features[key]["info"] = src_dataset.meta.info.features[key].get("info", {})
 
     write_info(dst_meta.info, dst_meta.root)
 
@@ -1269,11 +1281,7 @@ def _estimate_frame_size_via_calibration(
     episode_indices: list[int],
     temp_dir: Path,
     fps: int,
-    vcodec: str,
-    pix_fmt: str,
-    g: int,
-    crf: int,
-    fast_decode: int,
+    camera_encoder: VideoEncoderConfig,
     num_calibration_frames: int = 30,
 ) -> float:
     """Estimate MB per frame by encoding a small calibration sample.
@@ -1287,11 +1295,7 @@ def _estimate_frame_size_via_calibration(
         episode_indices: List of episode indices being processed.
         temp_dir: Temporary directory for calibration files.
         fps: Frames per second for video encoding.
-        vcodec: Video codec (libsvtav1, h264, hevc).
-        pix_fmt: Pixel format (yuv420p, etc.).
-        g: GOP size (group of pictures).
-        crf: Constant Rate Factor (quality).
-        fast_decode: Fast decode tuning parameter.
+        camera_encoder: Video encoder settings used for calibration encoding.
         num_calibration_frames: Number of frames to use for calibration (default: 30).
 
     Returns:
@@ -1327,11 +1331,7 @@ def _estimate_frame_size_via_calibration(
             imgs_dir=calibration_dir,
             video_path=calibration_video_path,
             fps=fps,
-            vcodec=vcodec,
-            pix_fmt=pix_fmt,
-            g=g,
-            crf=crf,
-            fast_decode=fast_decode,
+            camera_encoder=camera_encoder,
             overwrite=True,
         )
 
@@ -1525,7 +1525,7 @@ def modify_tasks(
     write_tasks(new_task_df, root)
 
     # Update info.json
-    dataset.meta.info["total_tasks"] = len(unique_tasks)
+    dataset.meta.info.total_tasks = len(unique_tasks)
     write_info(dataset.meta.info, root)
 
     # Reload metadata to reflect changes
@@ -1649,11 +1649,7 @@ def convert_image_to_video_dataset(
     dataset: LeRobotDataset,
     output_dir: Path | None = None,
     repo_id: str | None = None,
-    vcodec: str = "libsvtav1",
-    pix_fmt: str = "yuv420p",
-    g: int = 2,
-    crf: int = 30,
-    fast_decode: int = 0,
+    camera_encoder: VideoEncoderConfig | None = None,
     episode_indices: list[int] | None = None,
     num_workers: int = 4,
     max_episodes_per_batch: int | None = None,
@@ -1668,11 +1664,8 @@ def convert_image_to_video_dataset(
         dataset: The source LeRobot dataset with images
         output_dir: Root directory where the edited dataset will be stored. If not specified, defaults to $HF_LEROBOT_HOME/repo_id. Equivalent to new_root in EditDatasetConfig.
         repo_id: Edited dataset identifier. Equivalent to new_repo_id in EditDatasetConfig.
-        vcodec: Video codec (default: libsvtav1)
-        pix_fmt: Pixel format (default: yuv420p)
-        g: Group of pictures size (default: 2)
-        crf: Constant rate factor (default: 30)
-        fast_decode: Fast decode tuning (default: 0)
+        camera_encoder: Video encoder settings
+            (``None`` uses :func:`~lerobot.configs.camera_encoder_defaults`).
         episode_indices: List of episode indices to convert (None = all episodes)
         num_workers: Number of threads for parallel processing (default: 4)
         max_episodes_per_batch: Maximum episodes per video batch to avoid memory issues (None = no limit)
@@ -1681,6 +1674,9 @@ def convert_image_to_video_dataset(
     Returns:
         New LeRobotDataset with images encoded as videos
     """
+    if camera_encoder is None:
+        camera_encoder = camera_encoder_defaults()
+
     # Check that it's an image dataset
     if len(dataset.meta.video_keys) > 0:
         raise ValueError(
@@ -1704,7 +1700,10 @@ def convert_image_to_video_dataset(
     logging.info(
         f"Converting {len(episode_indices)} episodes with {len(img_keys)} cameras from {dataset.repo_id}"
     )
-    logging.info(f"Video codec: {vcodec}, pixel format: {pix_fmt}, GOP: {g}, CRF: {crf}")
+    logging.info(
+        f"Video codec: {camera_encoder.vcodec}, pixel format: {camera_encoder.pix_fmt}, "
+        f"GOP: {camera_encoder.g}, CRF: {camera_encoder.crf}"
+    )
 
     # Create new features dict, converting image features to video features
     new_features = {}
@@ -1774,11 +1773,7 @@ def convert_image_to_video_dataset(
                 episode_indices=episode_indices,
                 temp_dir=temp_dir,
                 fps=fps,
-                vcodec=vcodec,
-                pix_fmt=pix_fmt,
-                g=g,
-                crf=crf,
-                fast_decode=fast_decode,
+                camera_encoder=camera_encoder,
             )
 
             logging.info(f"Processing camera: {img_key}")
@@ -1820,11 +1815,7 @@ def convert_image_to_video_dataset(
                     imgs_dir=imgs_dir,
                     video_path=video_path,
                     fps=fps,
-                    vcodec=vcodec,
-                    pix_fmt=pix_fmt,
-                    g=g,
-                    crf=crf,
-                    fast_decode=fast_decode,
+                    camera_encoder=camera_encoder,
                     overwrite=True,
                 )
 
@@ -1858,10 +1849,10 @@ def convert_image_to_video_dataset(
         episodes_df.to_parquet(episodes_path, index=False)
 
         # Update metadata info
-        new_meta.info["total_episodes"] = len(episode_indices)
-        new_meta.info["total_frames"] = sum(ep["length"] for ep in all_episode_metadata.values())
-        new_meta.info["total_tasks"] = dataset.meta.total_tasks
-        new_meta.info["splits"] = {"train": f"0:{len(episode_indices)}"}
+        new_meta.info.total_episodes = len(episode_indices)
+        new_meta.info.total_frames = sum(ep["length"] for ep in all_episode_metadata.values())
+        new_meta.info.total_tasks = dataset.meta.total_tasks
+        new_meta.info.splits = {"train": f"0:{len(episode_indices)}"}
 
         # Update video info for all image keys (now videos)
         # We need to manually set video info since update_video_info() checks video_keys first
@@ -1870,7 +1861,9 @@ def convert_image_to_video_dataset(
                 video_path = new_meta.root / new_meta.video_path.format(
                     video_key=img_key, chunk_index=0, file_index=0
                 )
-                new_meta.info["features"][img_key]["info"] = get_video_info(video_path)
+                new_meta.info.features[img_key]["info"] = get_video_info(
+                    video_path, camera_encoder=camera_encoder
+                )
 
         write_info(new_meta.info, new_meta.root)
 
@@ -1893,3 +1886,83 @@ def convert_image_to_video_dataset(
 
     # Return new dataset
     return LeRobotDataset(repo_id=repo_id, root=output_dir)
+
+
+def _reencode_video_worker(args: tuple) -> Path:
+    """Picklable worker for :func:`reencode_dataset`'s process pool."""
+    video_path, camera_encoder, encoder_threads = args
+    reencode_video(
+        input_video_path=video_path,
+        output_video_path=video_path,
+        camera_encoder=camera_encoder,
+        encoder_threads=encoder_threads,
+        overwrite=True,
+    )
+    return video_path
+
+
+def reencode_dataset(
+    dataset: LeRobotDataset,
+    camera_encoder: VideoEncoderConfig,
+    encoder_threads: int | None = None,
+    num_workers: int | None = None,
+) -> LeRobotDataset:
+    """Re-encode every video in a dataset with a new set of encoding parameters.
+
+    Videos are re-encoded in-place and the video information in ``info.json`` is refreshed.
+
+    Args:
+        dataset: An existing :class:`LeRobotDataset` whose videos will be
+            re-encoded.
+        camera_encoder: Target encoder configuration applied to every video
+            file.
+        encoder_threads: Per-encoder thread count forwarded to
+            :func:`reencode_video`. ``None`` lets the codec decide.
+        num_workers: Number of parallel processes. ``None`` or ``0`` means
+            sequential (no multiprocessing); ``1+`` spawns a
+            :class:`~concurrent.futures.ProcessPoolExecutor`.
+
+    Returns:
+        The same :class:`LeRobotDataset` instance with its metadata updated
+        on disk.
+    """
+    meta = dataset.meta
+    video_paths_list = []
+
+    # Only re-encode if the videos are not already encoded with the given video encoding parameters
+    for video_key in meta.video_keys:
+        current_info = meta.info.features[video_key].get("info", {})
+        current_encoder = VideoEncoderConfig.from_video_info(current_info)
+        if current_encoder != camera_encoder:
+            video_paths_list.extend((meta.root / VIDEO_DIR / video_key).rglob("*.mp4"))
+        else:
+            logging.info(f"{video_key} videos are already encoded with {camera_encoder}. Nothing to do.")
+
+    if len(video_paths_list) == 0:
+        logging.warning("Dataset has no videos to re-encode.")
+        return dataset
+    logging.info(f"Re-encoding {len(video_paths_list)} video file(s) with {camera_encoder}")
+
+    worker_args = [(vp, camera_encoder, encoder_threads) for vp in video_paths_list]
+    if num_workers and num_workers > 1:
+        with ProcessPoolExecutor(max_workers=num_workers) as pool:
+            futures = [pool.submit(_reencode_video_worker, args) for args in worker_args]
+            for future in tqdm(
+                as_completed(futures),
+                total=len(futures),
+                desc="Re-encoding videos",
+            ):
+                future.result()
+    else:
+        for args in tqdm(worker_args, desc="Re-encoding videos"):
+            _reencode_video_worker(args)
+
+    # Refresh video info in metadata for every video key.
+    for vid_key in meta.video_keys:
+        video_path = meta.root / meta.get_video_file_path(0, vid_key)
+        meta.info.features[vid_key]["info"] = get_video_info(video_path, camera_encoder=camera_encoder)
+
+    write_info(meta.info, meta.root)
+    logging.info("Dataset metadata updated.")
+
+    return dataset

src/lerobot/datasets/dataset_writer.py

@@ -31,6 +31,8 @@ import PIL.Image
 import pyarrow.parquet as pq
 import torch
 
+from lerobot.configs import VideoEncoderConfig, camera_encoder_defaults
+
 from .compute_stats import compute_episode_stats
 from .dataset_metadata import LeRobotDatasetMetadata
 from .feature_utils import (
@@ -65,14 +67,19 @@ def _encode_video_worker(
     episode_index: int,
     root: Path,
     fps: int,
-    vcodec: str = "libsvtav1",
+    camera_encoder: VideoEncoderConfig | None = None,
     encoder_threads: int | None = None,
 ) -> Path:
     temp_path = Path(tempfile.mkdtemp(dir=root)) / f"{video_key}_{episode_index:03d}.mp4"
     fpath = DEFAULT_IMAGE_PATH.format(image_key=video_key, episode_index=episode_index, frame_index=0)
     img_dir = (root / fpath).parent
     encode_video_frames(
-        img_dir, temp_path, fps, vcodec=vcodec, overwrite=True, encoder_threads=encoder_threads
+        img_dir,
+        temp_path,
+        fps,
+        camera_encoder=camera_encoder,
+        encoder_threads=encoder_threads,
+        overwrite=True,
     )
     shutil.rmtree(img_dir)
     return temp_path
@@ -89,20 +96,22 @@ class DatasetWriter:
         self,
         meta: LeRobotDatasetMetadata,
         root: Path,
-        vcodec: str,
+        camera_encoder: VideoEncoderConfig | None,
         encoder_threads: int | None,
         batch_encoding_size: int,
         streaming_encoder: StreamingVideoEncoder | None = None,
         initial_frames: int = 0,
     ):
-        """Initialize the writer with metadata, codec, and encoding config.
+        """Initialize the writer with metadata, codec, and encoder config.
 
         Args:
             meta: Dataset metadata instance (used for feature schema, chunk
                 settings, and episode persistence).
             root: Local dataset root directory.
-            vcodec: Video codec for encoding (e.g. ``'libsvtav1'``, ``'h264'``).
-            encoder_threads: Threads per encoder instance. ``None`` for auto.
+            camera_encoder: Video encoder settings applied to all cameras.
+                ``None`` uses :func:`~lerobot.configs.camera_encoder_defaults`.
+            encoder_threads: Number of encoder threads (global). ``None``
+                lets the codec decide.
             batch_encoding_size: Number of episodes to accumulate before
                 batch-encoding videos.
             streaming_encoder: Optional pre-built :class:`StreamingVideoEncoder`
@@ -111,7 +120,7 @@ class DatasetWriter:
         """
         self._meta = meta
         self._root = root
-        self._vcodec = vcodec
+        self._camera_encoder = camera_encoder or camera_encoder_defaults()
         self._encoder_threads = encoder_threads
         self._batch_encoding_size = batch_encoding_size
         self._streaming_encoder = streaming_encoder
@@ -241,7 +250,14 @@ class DatasetWriter:
         for key, ft in self._meta.features.items():
             if key in ["index", "episode_index", "task_index"] or ft["dtype"] in ["image", "video"]:
                 continue
-            episode_buffer[key] = np.stack(episode_buffer[key])
+            stacked_values = np.stack(episode_buffer[key])
+
+            # `shape=(1,)` numeric features are serialized as `datasets.Value`, which expects scalars.
+            # Normalizing to `(N,)` keeps save semantics stable across dependency versions.
+            if tuple(ft["shape"]) == (1,) and ft["dtype"] != "string":
+                stacked_values = stacked_values.reshape(episode_length)
+
+            episode_buffer[key] = stacked_values
 
         # Wait for image writer to end, so that episode stats over images can be computed
         self._wait_image_writer()
@@ -284,7 +300,7 @@ class DatasetWriter:
                             episode_index,
                             self._root,
                             self._meta.fps,
-                            self._vcodec,
+                            self._camera_encoder,
                             self._encoder_threads,
                         ): video_key
                         for video_key in self._meta.video_keys
@@ -495,7 +511,7 @@ class DatasetWriter:
 
         # Update video info (only needed when first episode is encoded)
         if episode_index == 0:
-            self._meta.update_video_info(video_key)
+            self._meta.update_video_info(video_key, camera_encoder=self._camera_encoder)
             write_info(self._meta.info, self._meta.root)
 
         metadata = {
@@ -564,7 +580,12 @@ class DatasetWriter:
     def _encode_temporary_episode_video(self, video_key: str, episode_index: int) -> Path:
         """Use ffmpeg to convert frames stored as png into mp4 videos."""
         return _encode_video_worker(
-            video_key, episode_index, self._root, self._meta.fps, self._vcodec, self._encoder_threads
+            video_key,
+            episode_index,
+            self._root,
+            self._meta.fps,
+            self._camera_encoder,
+            self._encoder_threads,
         )
 
     def close_writer(self) -> None:

src/lerobot/datasets/factory.py

@@ -19,6 +19,7 @@ from pprint import pformat
 import torch
 
 from lerobot.configs import PreTrainedConfig
+from lerobot.configs.rewards import RewardModelConfig
 from lerobot.configs.train import TrainPipelineConfig
 from lerobot.transforms import ImageTransforms
 from lerobot.utils.constants import ACTION, IMAGENET_STATS, OBS_PREFIX, REWARD
@@ -30,12 +31,14 @@ from .streaming_dataset import StreamingLeRobotDataset
 
 
 def resolve_delta_timestamps(
-    cfg: PreTrainedConfig, ds_meta: LeRobotDatasetMetadata
+    cfg: PreTrainedConfig | RewardModelConfig, ds_meta: LeRobotDatasetMetadata
 ) -> dict[str, list] | None:
-    """Resolves delta_timestamps by reading from the 'delta_indices' properties of the PreTrainedConfig.
+    """Resolves delta_timestamps by reading from the 'delta_indices' properties of the config.
 
     Args:
-        cfg (PreTrainedConfig): The PreTrainedConfig to read delta_indices from.
+        cfg (PreTrainedConfig | RewardModelConfig): The config to read delta_indices from. Both
+            ``PreTrainedConfig`` and concrete ``RewardModelConfig`` subclasses expose the
+            ``{observation,action,reward}_delta_indices`` properties used below.
         ds_meta (LeRobotDatasetMetadata): The dataset from which features and fps are used to build
             delta_timestamps against.
 
@@ -82,7 +85,7 @@ def make_dataset(cfg: TrainPipelineConfig) -> LeRobotDataset | MultiLeRobotDatas
         ds_meta = LeRobotDatasetMetadata(
             cfg.dataset.repo_id, root=cfg.dataset.root, revision=cfg.dataset.revision
         )
-        delta_timestamps = resolve_delta_timestamps(cfg.policy, ds_meta)
+        delta_timestamps = resolve_delta_timestamps(cfg.trainable_config, ds_meta)
         if not cfg.dataset.streaming:
             dataset = LeRobotDataset(
                 cfg.dataset.repo_id,

src/lerobot/datasets/feature_utils.py

@@ -13,21 +13,30 @@
 # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and
 # limitations under the License.
+import logging
 from pprint import pformat
 
 import datasets
 import numpy as np
 from PIL import Image as PILImage
 
+from lerobot.configs import VIDEO_ENCODER_INFO_KEYS
 from lerobot.utils.constants import DEFAULT_FEATURES
 from lerobot.utils.utils import is_valid_numpy_dtype_string
 
+from .language import (
+    LANGUAGE_PERSISTENT,
+    is_language_column,
+    language_events_column_feature,
+    language_persistent_column_feature,
+)
 from .utils import (
     DEFAULT_CHUNK_SIZE,
     DEFAULT_DATA_FILE_SIZE_IN_MB,
     DEFAULT_DATA_PATH,
     DEFAULT_VIDEO_FILE_SIZE_IN_MB,
     DEFAULT_VIDEO_PATH,
+    DatasetInfo,
 )
 
 
@@ -45,7 +54,13 @@ def get_hf_features_from_features(features: dict) -> datasets.Features:
     """
     hf_features = {}
     for key, ft in features.items():
-        if ft["dtype"] == "video":
+        if is_language_column(key):
+            hf_features[key] = (
+                language_persistent_column_feature()
+                if key == LANGUAGE_PERSISTENT
+                else language_events_column_feature()
+            )
+        elif ft["dtype"] == "video":
             continue
         elif ft["dtype"] == "image":
             hf_features[key] = datasets.Image()
@@ -78,8 +93,8 @@ def create_empty_dataset_info(
     chunks_size: int | None = None,
     data_files_size_in_mb: int | None = None,
     video_files_size_in_mb: int | None = None,
-) -> dict:
-    """Create a template dictionary for a new dataset's `info.json`.
+) -> DatasetInfo:
+    """Create a template ``DatasetInfo`` object for a new dataset's ``meta/info.json``.
 
     Args:
         codebase_version (str): The version of the LeRobot codebase.
@@ -87,25 +102,59 @@ def create_empty_dataset_info(
         features (dict): The LeRobot features dictionary for the dataset.
         use_videos (bool): Whether the dataset will store videos.
         robot_type (str | None): The type of robot used, if any.
+        chunks_size (int | None): Max files per chunk directory. Defaults to ``DEFAULT_CHUNK_SIZE``.
+        data_files_size_in_mb (int | None): Max parquet file size in MB. Defaults to ``DEFAULT_DATA_FILE_SIZE_IN_MB``.
+        video_files_size_in_mb (int | None): Max video file size in MB. Defaults to ``DEFAULT_VIDEO_FILE_SIZE_IN_MB``.
 
     Returns:
-        dict: A dictionary with the initial dataset metadata.
+        DatasetInfo: A typed dataset information object with initial metadata.
     """
-    return {
-        "codebase_version": codebase_version,
-        "robot_type": robot_type,
-        "total_episodes": 0,
-        "total_frames": 0,
-        "total_tasks": 0,
-        "chunks_size": chunks_size or DEFAULT_CHUNK_SIZE,
-        "data_files_size_in_mb": data_files_size_in_mb or DEFAULT_DATA_FILE_SIZE_IN_MB,
-        "video_files_size_in_mb": video_files_size_in_mb or DEFAULT_VIDEO_FILE_SIZE_IN_MB,
-        "fps": fps,
-        "splits": {},
-        "data_path": DEFAULT_DATA_PATH,
-        "video_path": DEFAULT_VIDEO_PATH if use_videos else None,
-        "features": features,
-    }
+    return DatasetInfo(
+        codebase_version=codebase_version,
+        fps=fps,
+        features=features,
+        robot_type=robot_type,
+        chunks_size=chunks_size or DEFAULT_CHUNK_SIZE,
+        data_files_size_in_mb=data_files_size_in_mb or DEFAULT_DATA_FILE_SIZE_IN_MB,
+        video_files_size_in_mb=video_files_size_in_mb or DEFAULT_VIDEO_FILE_SIZE_IN_MB,
+        data_path=DEFAULT_DATA_PATH,
+        video_path=DEFAULT_VIDEO_PATH if use_videos else None,
+    )
+
+
+def features_equal_for_merge(features_a: dict[str, dict], features_b: dict[str, dict]) -> bool:
+    """Return whether two LeRobotDatasetMetadata ``features`` dicts are compatible for aggregation.
+
+    For video features, keys under ``info`` related to video encoding parameters are ignored during
+    comparison as they do not prevent aggregation.
+    """
+
+    def _without_encoder_info_keys(feature: dict) -> dict:
+        filtered = dict(feature)
+        filtered_info = filtered.get("info")
+        if isinstance(filtered_info, dict):
+            filtered["info"] = {
+                info_key: info_value
+                for info_key, info_value in filtered_info.items()
+                if info_key not in VIDEO_ENCODER_INFO_KEYS
+            }
+        return filtered
+
+    if set(features_a) != set(features_b):
+        return False
+    for key in features_a:
+        fa_key = features_a[key]
+        fb_key = features_b[key]
+        if fa_key.get("dtype") != fb_key.get("dtype"):
+            return False
+        if fa_key.get("dtype") != "video":
+            if fa_key != fb_key:
+                return False
+            continue
+
+        if _without_encoder_info_keys(fa_key) != _without_encoder_info_keys(fb_key):
+            return False
+    return True
 
 
 def check_delta_timestamps(
@@ -242,6 +291,8 @@ def validate_feature_dtype_and_shape(
         return validate_feature_image_or_video(name, expected_shape, value)
     elif expected_dtype == "string":
         return validate_feature_string(name, value)
+    elif expected_dtype == "language":
+        return validate_feature_language(name, value)
     else:
         raise NotImplementedError(f"The feature dtype '{expected_dtype}' is not implemented yet.")
 
@@ -321,6 +372,30 @@ def validate_feature_string(name: str, value: str) -> str:
     return ""
 
 
+def validate_feature_language(name: str, value) -> str:
+    """Validate a feature that is expected to hold language annotations.
+
+    Language columns (``language_persistent`` / ``language_events``) are
+    populated after recording by the annotation pipeline, not at record time.
+    Any value supplied here is dropped before the frame is written, so a
+    non-empty value almost certainly signals a mistake. We warn rather than
+    fail to keep recording resilient.
+
+    Args:
+        name (str): The name of the feature.
+        value: The value to validate.
+
+    Returns:
+        str: Always an empty string — language values are non-fatal.
+    """
+    if value is not None:
+        logging.warning(
+            f"The feature '{name}' is a 'language' column populated by the annotation pipeline, "
+            f"not at record time. The provided value will be dropped."
+        )
+    return ""
+
+
 def validate_episode_buffer(episode_buffer: dict, total_episodes: int, features: dict) -> None:
     """Validate the episode buffer before it's written to disk.