Reproduce “OneSteer: Unlocking Steerable General Humanoid Motion Tracking via One-Step Diffusion Whole-Body Control” 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

Yifei Huang, Yazhao Yang, Li Weijia, Tianle Shi, Ji Jin, Zhihui Cai, Zihao Liu, Jianzhuang Zhao and 5 more

OneSteer steers a humanoid robot’s predicted motion toward task objectives without changing weights. It optimizes generator input using a differentiable cost, then a feedback-conditioned action head executes a command and replans; masks can relax conflicting reference groups. In simulation, it achieved a 0.90 clean-pass rate under a 1.25 m bar, versus at most 0.11 for tested editors, across 79 windows.

Summary by alphaXiv.

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