From 1f74982469d139bbb9cf93df8c14f27aeb16c2f9 Mon Sep 17 00:00:00 2001 From: Bryson Jones Date: Thu, 11 Dec 2025 09:55:41 -0800 Subject: [PATCH] note origins of each training objective --- docs/source/multitask_dit.mdx | 6 +++++- 1 file changed, 5 insertions(+), 1 deletion(-) diff --git a/docs/source/multitask_dit.mdx b/docs/source/multitask_dit.mdx index 3bb1095d7..97a1864e3 100644 --- a/docs/source/multitask_dit.mdx +++ b/docs/source/multitask_dit.mdx @@ -187,7 +187,11 @@ The vision encoder uses a separate learning rate multiplier, where 1/10th is sug #### 1. Flow Matching with Beta Sampling -Consider switching to flow matching with beta sampling distribution for potentially improved performance: +The original diffusion implementation here is based on the work described in [TRI's LBM paper](https://arxiv.org/abs/2507.05331) + +Additionally, we have implemented a flow-matching objective, which is described at a high-level in [Boston Dynamics blog post](https://bostondynamics.com/blog/large-behavior-models-atlas-find-new-footing/). + +Consider testing the flow-matching objective and evaluating performance differences for your task: ```bash --policy.objective=flow_matching \