Reproduce “P-BFM: Perceptive Behavioral Foundation Models for Humanoid Control via Unsupervised Reinforcement Learning” 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
Xue Wang, Yitang Li, Yiming Zhang, Siqin Wang, Junhan Sun, Zhihao Liu, Yaojie Shen, Jingbo Yuan and 2 more
P-BFM separates behavior intent from terrain-specific execution, using depth perception and explicit heading and velocity instructions to adapt movement. It learns through interaction with varied simulated terrain, using 40 flat-ground LaFAN motion sequences as prompts rather than terrain-matched demonstrations. On unseen SOMA motions, it achieves 6.7 ± 4.7° mean root-heading error and 61.4% recovery rate in simulation.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.