Reproduce “CLOT: Closed-Loop Global Motion Tracking for Whole-Body Humanoid Teleoperation” 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
Tengjie Zhu, Guanyu Cai, Yang Zhaohui, Guanzhu Ren, Haohui Xie, ZiRui Wang, Junsong Wu, Jingbo Wang and 3 more
A closed-loop global motion tracking framework, CLOT, enables drift-free whole-body humanoid teleoperation by integrating high-frequency localization feedback for long-horizon tasks. This system from Shanghai Jiao Tong University and Shanghai AI Lab shows an order of magnitude reduction in tracking errors compared to prior methods, coupled with robust performance on a full-sized humanoid.
Key insights
- Directly addressing global pose drift through closed-loop feedback allows for stable, long-horizon humanoid teleoperation, crucial for practical applications.
- The "Observation Pre-shift" technique effectively teaches RL policies to perform smooth global error corrections by implicitly learning compliant motion interpolation.
- A Transformer architecture is vital for learning long-horizon spatiotemporal dependencies, enabling robust tracking and leveraging strategies like Observation Pre-shift.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.