Reproduce “Learning Agile Humanoid Dodgeball with Onboard Monocular Vision” 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

Feng Tian, Yulin An, Chenhao Li, Matthias Heyrman, Marco Hutter

The system combines monocular ball detection and temporal position/threat estimation with learned dodge-recovery and search-tracking policies for humanoid dodgeball. It trains control with ground-truth ball states before switching to estimated states while estimator learning continues. In simulation, center-bin success is 100.0% at 6 m/s and 5 m, versus 39.9% at 10 m/s and 2.5 m, demonstrating performance only under tested ball and launch conditions.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.