admin/lerobot
1
0
Fork 0
mirror of https://github.com/huggingface/lerobot.git synced 2026-06-27 05:07:15 +00:00
Code Issues Packages Projects Releases Wiki Activity

fix(evo1): move LIBERO padding into policy processors

Browse Source
This commit is contained in:
javadcc_mac
2026-06-21 15:58:38 +08:00
parent 4cfa762da8
commit 25556ceefe
16 changed files with 637 additions and 252 deletions
Download Patch File Download Diff File Expand all files Collapse all files
docs/source/evo1.mdx
+9 -3
View File
@@ -139,6 +139,8 @@ every finetuning flag.
| `policy.n_action_steps` | `50` | Number of actions consumed from a sampled chunk |
| `policy.max_state_dim` | `24` | State padding dimension |
| `policy.max_action_dim` | `24` | Action padding dimension |
| `policy.postprocess_action_dim` | `null` | Optional action dimension returned after EVO1 postprocessing |
| `policy.binarize_gripper` | `false` | Binarizes the postprocessed gripper channel for LIBERO-style eval |
| `policy.task_field` | `task` | Batch field used as the language prompt |
## Results
@@ -161,16 +163,20 @@ pixel embeddings, VLM fused tokens, normalized actions, and denormalized actions
The published checkpoint expects the raw LIBERO camera feature names
`observation.images.agentview_image` and `observation.images.robot0_eye_in_hand_image`. The official EVO1 LIBERO
rollout protocol also replans every 14 actions and binarizes the gripper command before stepping the simulator.
The LIBERO environment postprocessor applies the gripper binarization automatically for EVO1 policies. To run the
converted checkpoint with LeRobot LIBERO evaluation for the same one-episode-per-task setting, keep the raw camera
names instead of the default `image`/`image2` mapping and override `policy.n_action_steps` to 14:
The EVO1 policy postprocessor can crop the padded 24D action back to the 7D LIBERO action space and apply that
gripper binarization. To run the converted checkpoint with LeRobot LIBERO evaluation for the same
one-episode-per-task setting, keep the raw camera names instead of the default `image`/`image2` mapping, enable
FlashAttention, and set the LIBERO action postprocessing flags:
```bash
lerobot-eval \
--policy.path=javadcc/evo1-libero-lerobot \
--policy.vlm_model_name=OpenGVLab/InternVL3-1B \
--policy.device=cuda \
--policy.use_flash_attn=true \
--policy.n_action_steps=14 \
--policy.postprocess_action_dim=7 \
--policy.binarize_gripper=true \
--env.type=libero \
--env.task=libero_object \
--env.camera_name_mapping="{agentview_image: agentview_image, robot0_eye_in_hand_image: robot0_eye_in_hand_image}" \