Reproduce “It Takes Two: Learning Interactive Whole-Body Control Between Humanoid Robots” 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
Zuhong Liu, Junhao Ge, Minhao Xiong, Jiahao Gu, Bowei Tang, Wei Jing, Siheng Chen
Researchers at Shanghai Jiao Tong University developed Harmanoid, a framework enabling two humanoid robots to perform physically grounded, interactive whole-body motions. It achieves a 0.92 success rate in complex interactive tasks, significantly outperforming single-robot imitation methods.
Key insights
- Treating humanoid robots independently for interactive tasks leads to catastrophic failures, highlighting the necessity of explicitly modeling inter-agent dynamics.
- A systematic approach that first generates contact-aware, interaction-preserving reference motions via retargeting and then refines them with an interaction-driven controller is more effective.
- Integrating interaction-aware rewards and real-time contact feedback within a reinforcement learning framework enables robots to learn physically plausible and coordinated interactive behaviors.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.