Reproduce “Humanoid Dodgeball: Learning Fast Dynamic Obstacle Avoidance Motions for Humanoids” 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
Aoqian Zhang, Chunzheng Wang, Haoran Chen, Annan Tang, Cheng Xiang, Fan Shi, Shuzhi Sam Ge
Humanoid Dodgeball trains a humanoid robot to evade thrown balls while maintaining balance, using threat-aware rewards, direction-specific motion priors, and capture-point guidance for recovery. In PM01 simulation, it achieved 94.2% success versus 79.5% for PAC-MAN in the 0.8–1.2-second nominal time-to-hit bin. On hardware, it dodged 22 of 30 trials using motion-capture obstacle positions rather than onboard perception.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.