Reproduce “PAC-MAN: Perception-Aware CBF-RL for Whole-Body Safety in Humanoid Dodgeball” 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
Lizhi Yang, Junheng Li, Aaron D. Ames
PAC-MAN trains a humanoid to evade moving balls using safety guidance for each body link, then deploys a policy driven by camera-derived depth and proprioception alone. Link-CBF guidance better fits fixed-camera observations than Joint-CBF guidance in simulation: deployment success was 89% versus 76%. On a Unitree G1, it dodged 19 of 20 throws without falling, in a small hardware test.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.