Reproduce “Test-Time Compute for Humanoid Behavior Foundation Models” 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
Weiji Xie, Junsong Wu, Zibo Zhou, Yueru Chen, Xingyi Wang, Zhenhua Xiong, Weinan Zhang, Chenjia Bai
TTCO optimizes motion commands to improve how frozen humanoid behavior foundation models track desired robot motions. It searches candidate commands by simulating their execution and selecting inputs that reduce trajectory error. In simulation across 12,273 motions, it lowers mean body-position error by 79.1–84.2% across three models at K=640, while increasing joint acceleration; the method is not yet fast enough for satisfactory real-time use.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.