Reproduce “WholeBodyWAM: Learning Whole-Body World Action Models with Scalable Motion Priors” 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
Bowei Zhang, Qiyao Zhang, Shuanghao Bai, Xinhua Wang, Meng Li, Yilei Wang, Leiwang Zhang, Jian Tang and 3 more
WholeBodyWAM trains a language-conditioned Motion Expert on heterogeneous human and robot motion to predict future articulated body movement, then connects it with visual and action experts for robot-specific control. Its 63D representation excludes global root movement, and future visual tokens are omitted at deployment. On six TianGong 3.0 tasks, it averages 72.2% versus 60.8%, scores rather than binary success rates.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.