Implicit Kinodynamic Motion Retargeting for Human-to-humanoid Imitation Learning

Xingyu Chen1, Hanyu Wu2, Sikai Wu1 Mingliang Zhou1 Diyun Xiang1 Haodong Zhang3
1. Xiaomi Robotics Lab, 2. ETH Zurich, 3. Zhejiang University
arXiv Paper Code

Overview

A real-time method for retargeting large-scale human motion into robot-executable motion.

Third research result visualization

Retargeting Performance

By neural retargeting, We achieve processing speed >5000 fps

Retargeting Details

IKMR converts the human motion into robotics by skeleton features

Comparison

We compaire IKMR with common methods used in whole-body control

Real Robot

The final performance in the real robotics

Video Presentation

BibTeX


@article{Chen2025ImplicitKM,
  title={Implicit Kinodynamic Motion Retargeting for Human-to-humanoid Imitation Learning},
  author={Xingyu Chen and Hanyu Wu and Sikai Wu and Mingliang Zhou and Diyun Xiang and Haodong Zhang},
  journal={arXiv preprint arXiv:2509.15443},
  year={2025},
  url={https://arxiv.org/abs/2509.15443}
}