anonymised-submission/SO101_Toy2Can

SO101 Toy2Can

LeRobot demonstration dataset collected with an SO-101 follower arm for tabletop pick-and-place into a trash can. It is released under an anonymous submission account to document ongoing real-hardware work associated with a simulation-primary workshop submission.

Dataset summary

• 180 episodes and 40,910 frames at 30 FPS.
• Six-dimensional joint-position observations and actions.
• Two 640 x 480 RGB video streams: a side camera (observation.images.SideLeft) and a wrist camera (observation.images.Wrist).
• 100 episodes: place a toy popcorn bucket in the can.
• 80 episodes: place a toy ketchup bottle in the can.
• Robot type recorded in the LeRobot metadata: so101_follower.

The companion target-robot demonstrations are available in XLERobot Toy2Can 640x480.

Scope and limitations

These are demonstration trajectories, not a protocol-controlled evaluation set. The release does not provide checkpoint-linked closed-loop success labels and should not be interpreted as evidence of real-robot policy performance or sim-to-real transfer. Camera views may contain ordinary laboratory background details.

Loading with LeRobot

from lerobot.datasets.lerobot_dataset import LeRobotDataset

dataset = LeRobotDataset("anonymised-submission/SO101_Toy2Can")

License

Apache-2.0, matching the source dataset card.