Reproduce “MoViC: Real-Time Humanoid Whole-Body Control from Monocular Video Streams” 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

Yimin Han, Jiahui Zhang, Wanyue Li, Zeren Luo, Yeke Chen, Liu Zhao, Peizhuo Li, Xiaoyu Ji and 5 more

MoViC turns monocular RGB video into humanoid robot references, retargeting human motion and extrapolating between visual updates for a 50-Hz controller. Its policy combines one next-step target with compact frequency-domain motion context, θ. In an ablation, this input used 1,182 features and had 99% success, versus 95% with ten explicit future frames. This supports compactness, not runtime savings or general performance.

Summary by alphaXiv.

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