ydy0615/lerobot-clone
1
0
Fork 0
mirror of https://github.com/huggingface/lerobot.git synced 2026-06-01 19:31:25 +00:00
Code Issues Packages Projects Releases Wiki Activity

modifications to gym_manipulator and buffer

Browse Source
This commit is contained in:
Michel Aractingi
2025-04-07 08:45:53 +02:00
parent ab2c2d39fb
commit f3cea2a3e5
8 changed files with 76 additions and 78 deletions
Download Patch File Download Diff File Expand all files Collapse all files

30
lerobot/scripts/server/actor_server.py
View File

@@ -250,28 +250,18 @@ def act_with_policy(
logging.info("[ACTOR] Shutting down act_with_policy")
return
if interaction_step >= cfg.policy.online_step_before_learning:
# Time policy inference and check if it meets FPS requirement
with TimerManager(
elapsed_time_list=list_policy_time,
label="Policy inference time",
log=False,
) as timer: # noqa: F841
action = policy.select_action(batch=obs)
policy_fps = 1.0 / (list_policy_time[-1] + 1e-9)
# Time policy inference and check if it meets FPS requirement
with TimerManager(
elapsed_time_list=list_policy_time,
label="Policy inference time",
log=False,
) as timer: # noqa: F841
action = policy.select_action(batch=obs)
policy_fps = 1.0 / (list_policy_time[-1] + 1e-9)
log_policy_frequency_issue(policy_fps=policy_fps, cfg=cfg, interaction_step=interaction_step)
log_policy_frequency_issue(policy_fps=policy_fps, cfg=cfg, interaction_step=interaction_step)
next_obs, reward, done, truncated, info = online_env.step(action.squeeze(dim=0).cpu().numpy())
else:
# TODO (azouitine): Make a custom space for torch tensor
action = online_env.action_space.sample()
next_obs, reward, done, truncated, info = online_env.step(action)
# HACK: We have only one env but we want to batch it, it will be resolved with the torch box
action = (
torch.from_numpy(action[0]).to(device, non_blocking=device.type == "cuda").unsqueeze(dim=0)
)
next_obs, reward, done, truncated, info = online_env.step(action.squeeze(dim=0).cpu().numpy())
sum_reward_episode += float(reward)
# Increment total steps counter for intervention rate