Research Sources

Citing this work

These hardware designs were developed for, and validated in, GR00T 2. If you use GEAR DexUMI or the other GR00T Data Hardware Prototypes in your work, please cite the GR00T 2 paper. Please also cite the original DexUMI paper below, which GEAR DexUMI builds on.

GR00T 2: Scaling Robot Foundation Models for Generalization, Dexterity, and Adaptation

NVIDIA GEAR.

GR00T 2 page

@article{gr00t2_2026,
  title   = {GR00T 2: Scaling Robot Foundation Models for
             Generalization, Dexterity, and Adaptation},
  author  = {{NVIDIA}},
  year    = {2026},
  url     = {https://research.nvidia.com/labs/gear/gr00t-2/}
}

Papers

DexUMI: Using Human Hand as the Universal Manipulation Interface for Dexterous Manipulation

Conference on Robot Learning (CoRL) 2025.

Presents a framework that collects demonstrations with a human hand wearing a hand exoskeleton and transfers the skills to different robot hands. The exoskeleton reduces the kinematic gap between human and robot hands and provides haptic feedback, while video inpainting replaces the human hand with the robot hand to reduce the visual gap.

Project page · Paper (arXiv) · Code

@inproceedings{xu2025dexumi,
  title={DexUMI: Using Human Hand as the Universal Manipulation Interface for Dexterous Manipulation},
  author={Xu, Mengda and Zhang, Han and Hou, Yifan and Xu, Zhenjia and Fan, Linxi and Veloso, Manuela and Song, Shuran},
  booktitle={Conference on Robot Learning},
  pages={437--459},
  year={2025},
  organization={PMLR}
}