Reproduce “Action Intent as a Bridge: Transferring Manipulation Knowledge from Human-Centric Videos to Robots” 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
Zhiqian Xia, Lianyang Ma, Haifeng Xia, Siyu Xia, Wei Wang, Xiaochun Cao
The authors introduce AIT to transfer human manipulation videos to robot learning. It represents actions as ordered sequences of coarse motion primitives, pairs them with learned implicit intent for finer motion, and decodes this representation into robot commands. On a 12-task Astribot S1 evaluation, AIT reports 82.08% average success versus UniVLA’s 70.42%, with both models task-fine-tuned on robot data.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.