Reproduce “AIRoA-0: Long-Horizon, Precise, Cross-Embodiment Manipulation from Handheld Data” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Yueh-Hua Wu, Kei Ota, Hanlong Li, Hayata Yuasa, Jason Wang, Kengo Ikeuchi, Kensuke Chiyoma, Makoto Sugiura and 12 more
AIRoA-0 trains robot policies from handheld demonstrations recorded with a gripper carrying the robot’s wrist camera, without collecting robot data. Its system predicts task stages and streams gripper-relative actions from two wrist-camera views. On ten bimanual Franka tasks at an unseen site, it averaged 78% success, versus 27% for π0.5; this result covers those tasks and conditions.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.