Reproduce “Zero-HLM: Generalizable Humanoid Loco-Manipulation without Real Robot Data” 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
Dianyi Yang, Mengya Liu, Zhumei Wang, Yu Liu, Xiaochen Zhang, Yixuan Li, Le Ma, Yuhang Yang and 4 more
Zero-HLM generates humanoid simulation training data by reconstructing scenes from smartphone captures, adapting demonstrations, and correcting motions through simulated execution. It produces 4,800 validated demonstrations from 45 source demonstrations. On a Unitree G1, GR00T N1.7's mean full-task success is 66.7% with 800 simulated demonstrations per task, versus 46.7% with 100 real demonstrations; evidence covers three tasks and tested configurations.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.