delta: fan-grasp policy checkpoints
LeRobot checkpoints for the Delta robot's fan grasp (left arm, wrist camera), used by the ROS 2 node
in https://github.com/s514041b10841041-bot/delta_policy_inference. Download into that repo's
checkpoints/ folder and check the files:
hf download ArthurGoldFishKing/delta_all_policy --local-dir checkpoints
cd checkpoints && sha256sum -c SHA256SUMS
| Folder | Policy name (--policy-name) |
Inputs | Steps |
|---|---|---|---|
diffusion/checkpoint_diffusion_375 |
diffusion_rtc / diffusion (default) |
wrist RGB + state | 375k |
diffusion/checkpoint_diffusion_50 |
same, --checkpoint diffusion_50k |
wrist RGB + state | 50k |
diffusion_depth/checkpoint_diffusion_375 |
depth_inference_rtc / depth_inference (default) |
RGB + depth + state | 375k |
diffusion_depth/checkpoint_diffusion_50 |
same, --checkpoint diffusion_50k |
RGB + depth + state | 50k |
diffusion_depth_only/checkpoint_diffusion_10 |
diffusion_depth_only_rtc / diffusion_depth_only |
depth + state | 10k |
act/checkpoint_act_375 |
act |
wrist RGB + state | 375k |
act_depth/checkpoint_act_375 |
depth_act |
RGB + depth + state | 375k |
Actions: 7-D per-step deltas at 30 Hz: position (m, world frame), rotation vector (rad, world
frame, left-multiplied), gripper (1 = open, 0 = closed), for the TCP at link6 + 0.2204 m. Depth
checkpoints include depth_encoding.json. Loaded with lerobot 0.6.2. None has been run on the
real robot yet.