Reproduce “XMimic and XSoma: Learning Humanoid Motion Tracking and Locomotion in Under a Minute” 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

Nico Bohlinger, Jan Peters

XMimic and XSoma speed up simulation-based humanoid training by compiling the training loop, shortening serial computation, and overlapping work across GPUs; XSoma also computes locomotion rewards directly from imagined states and terrain. On four H100 GPUs, they reach motion-tracking and locomotion targets in 0.86 and 0.77 minutes, respectively. These training times exclude startup and compilation, and results are simulation-only.

Summary by alphaXiv.

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