Ajingu/so101-imitation-learning-tape-dispenser

SO-101 ACT Policy: Tape Dispenser Pick-and-Place

ACT policy trained with LeRobot for a real SO-ARM101/SO-101 leader/follower setup.

• —Task: pick-and-place of a handheld tape dispenser
• —Training data: 30 teleoperated episodes
• —Observations: follower joint state + two externally installed USB camera views
• —Evaluation: 5/5 successful rollouts in the same fixed tabletop setup

This is a fixed-setup experiment artifact, not a robustness benchmark.

Demo

The video contains all five evaluation rollouts. In the first rollout, I accidentally reset the target object before stopping the episode, and the policy responded by picking it up and moving it back to the target area. Open the evaluation video <!-- <video src="./side-5evals.mp4" controls></video> --> <video controls src="https://huggingface.co/Ajingu/so101-imitation-learning-tape-dispenser/blob/main/side-5evals.mp4"></video>

Code

https://github.com/ajingu/Imitation-Learning_SO-101