Reproduce “ManiUnit: A Manipulation Skill Dataset and Benchmark for Long-Horizon Tasks” 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
Guoting Wei, Dawei Yan, Xia Yuan, Gengming Zhang, Yelin He, Guodong Du, Jiaquan Ye, Heng Zhang and 5 more
ManiUnit turns mobile household demonstrations across 50 activities into instructed manipulation segments and evaluates each operation from a restored starting state with its own local goal. In two simulated activities, skill-level training achieved 78.7% local success, compared with 49.3% for task-level training; a planner-coordinated system raised full-task success from 4.0% to 18.0%. These results do not establish performance beyond those activities.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.