Reproduce “VidAct: Learning Manipulation from In-the-Wild Videos with Object-Centric 3D Awareness” 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
Hang Li, Mingxin Zhang, Zihan Wu, Yang Tian, Dong Chen, Fengyi Shen, Yuan Meng, Xiangtong Yao and 4 more
VidAct converts monocular object-manipulation video into simulated robot demonstrations, then trains an RGB-only policy with auxiliary prediction of complete object point clouds. It transfers object motion relative to a static object rather than copying hand motion; 3D supervision is training-only. In unseen scenes, it improves real-world success by 11.67 percentage points over pose-only supervision across four tasks, for rigid-object manipulation.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.