SO-101 move cube (ACT policy evaluation)
Three autonomous evaluation episodes of an ACT policy driving an SO-101 arm to pick up a block and place it on a blue square. The policy was trained on only five demonstrations (lerobot_train_move_cube), my first policy for the SO-101. It succeeded on 1 of 3 episodes, which was surprising given such a small training set.
Contents
- Robot:
so101_follower - Episodes: 3
- Frames: 1,466
- FPS: 30
| Feature | Type | Details |
|---|---|---|
action |
float32[6] | shoulder_pan.pos, shoulder_lift.pos, elbow_flex.pos, wrist_flex.pos, wrist_roll.pos, gripper.pos |
observation.state |
float32[6] | shoulder_pan.pos, shoulder_lift.pos, elbow_flex.pos, wrist_flex.pos, wrist_roll.pos, gripper.pos |
observation.images.wrist |
video, 640x480 @ 30 fps |
This dataset is in LeRobot format (codebase version v3.0); the full schema is in meta/info.json.
Usage
from lerobot.datasets.lerobot_dataset import LeRobotDataset
ds = LeRobotDataset("primordial-spork/lerobot_eval_move_cube")
- Downloads last month
- 70