dobri420/pick-cube-so101

Pick-Cube SO-101

A deliberately structured teleoperated pick-and-place dataset on the SO-101 arm — 340 episodes of "grab the cube and put it in the box", recorded to probe how grasp coverage, reach depth, and wrist angle drive an imitation policy.

Composition

Five tranches, recorded in order:

Episodes	Tranche	What
0–49	discrete @ 25 cm	5 target angles (0°, ±45°, ±90°) × 10 demos
50–69	dense jitter	20 demos clustered at 45° right, ±2 cm placement jitter
70–119	discrete @ 15 cm	the same 5 angles at the near depth × 10 demos
120–239	blanket jitter	120 demos spread across the workspace
240–339	twist-CC jitter	100 demos with a counter-clockwise–twisted wrist for grasp-angle variability — not yet used in training

Grasp distribution

Each arrow is one episode's grasp — its position on the workspace half-disk (rings = reach in cm from the shoulder-pan axis, radials = target angle) and its wrist yaw (arrowhead points toward the upper jaw), recovered by SO-101 forward kinematics on the jaw center. The grid is calibrated from this dataset's own registration clusters.

The discrete tranches sit as tight clusters on the 15/25 cm rings; blanket jitter fills the gaps; twist-CC visibly rotates the arrows off-radial — the wrist variability the earlier tranches lack.

Recording

• Robot: SO-101 leader→follower pair (baby_gewu), teleoperated at 30 fps.
• Cameras: three views — camera1 top, camera2 wrist, camera3 side (480×640).
• State / action: 6-DoF joint positions (shoulder_pan … gripper), in degrees.
• Task: "Grab the cube and put it in the box."

Usage

from lerobot.datasets.lerobot_dataset import LeRobotDataset

ds = LeRobotDataset("dobri420/pick-cube-so101")
print(ds.num_episodes, ds.num_frames)  # 340 ...

Built with LeRobot (codebase_version: v3.0).