Reproduce “HuBE: Cross-Embodiment Human-like Behavior Execution for 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
Shipeng Lyu, Fangyuan Wang, Weiwei Lin, Luhao Zhu, David Navarro-Alarcon, Guodong Guo
Researchers from The Hong Kong Polytechnic University and Eastern Institute of Technology developed HuBE, a bi-level closed-loop framework for humanoid robots that generates behaviors exhibiting both kinematic similarity to human motion and contextual appropriateness. The system achieves millimeter-level accuracy when transferring behaviors across diverse robot morphologies and demonstrates improved computational efficiency compared to existing methods.
Key insights
- Explicitly integrating fine-grained, language-based situational contexts significantly enhances both the kinematic similarity and contextual appropriateness of generated robot behaviors.
- A bone scaling-based data augmentation strategy is effective for achieving millimeter-level cross-embodiment compatibility, enabling a single policy to generalize across heterogeneous humanoid robot designs.
- Hindsight training, by sampling future poses as targets, substantially enriches the training dataset and improves the model's ability to generate diverse and contextually appropriate human-like motions.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.