Does Diffusion Policy run on Unitree G1?

Capabilities

Caveats

• Documented by Unitree in its unitree_IL_lerobot repository, which pins LeRobot as a git submodule (commit a5b29d4, November 2025) and deploys with its own eval_robot/eval_g1.py script, not through LeRobot's native unitree_g1 robot or lerobot-rollout.
• The documented workflow covers the G1's arms and dexterous hands (--arm G1_29/G1_23, --ee dex1/dex3/inspire1/brainco); it does not document leg or locomotion control.

Evidence 2 sources; the cell shows the highest tier

1. T1 · upstream documentedvendor-docsverified 2026-10-24 (1d ago)lerobot a5b29d4

   Unitree's README gives a lerobot_train.py command for Diffusion Policy (--policy.type=diffusion) on its G1 Dex3 dataset.

   “--dataset.repo_id=unitreerobotics/G1_Dex3_ToastedBread_Dataset \ --policy.push_to_hub=false \ --policy.type=diffusion”

   https://github.com/unitreerobotics/unitree_IL_lerobot/blob/41c2805742de879ddab2d8d6beaeaf215f876395/README.md
2. T1 · upstream documentedvendor-docsverified 2026-10-24 (1d ago)lerobot a5b29d4

   Unitree documents real-robot testing of a trained LeRobot policy on the G1 via eval_g1.py (its example checkpoint path is a diffusion run).

   “To test your trained model on a real robot, you can use the eval_g1.py script located in the eval_robot folder.”

   https://github.com/unitreerobotics/unitree_IL_lerobot/blob/41c2805742de879ddab2d8d6beaeaf215f876395/README.md