Abstract

Robot manipulation policies are typically tied to specific robotic hand embodiments, limiting the transfer of learned behaviors across platforms with different kinematic structures. In this work, we propose the Unified Hand Action Space (UHAS), a sphere-based unified action representation for cross- embodiment dexterous manipulation. UHAS represents robotic hand actions as geometric deformations of a canonical sphere and uses a Cascade Inverse Kine- matics (CIK) algorithm to map the shared representation to embodiment-specific joint configurations. Using reinforcement learning, we train dexterous manipula- tion policies directly in the proposed action space for in-hand cube reorientation tasks. We evaluate our method in both simulation and real-world experiments across multiple robotic hands, including the Allegro Hand, LEAP Hand, Shadow Hand, and MANO Human Hand. Experimental results demonstrate effective dex- terous manipulation, zero-shot transfer to unseen hands, rapid finetuning across embodiments, and successful real-world deployment. Our experiments show that the proposed UHAS representation enables stable dexterous control and cross- embodiment policy transfer across robotic hands.

Overview

Overview our VLA-Replica benchmark

In our unified hand action space, an action is represented as the deformation of a canonical sphere. A deformed sphere is mapped to hand configurations of various embodiments (LEAP, Allegro, MANO Human and Shadow).

Automatic Sphere Creation for a Robotic Hand

Sphere Creation

Illustration of the process of creating a sphere for a robotic hand given its URDF.

Unified Hand Surface Correspondence

Sphere Creation

Construction of the unified hand surface correspondence.

Sphere Deformation Action Space

Sphere Creation

Sphere deformation parameterization in the Unified Hand Action Space (UHAS). (a) Initial configuration of four driving planes. (b) Rotating the driving planes controls the lateral deformation Δθ. (c) Radial displacement of the driving vectors controls Δr. (d) The final deformed sphere reconstructed through interpolation.

Cascade Inverse Kinematics

Sphere Creation

We classify hand joints into (a) lateral joints and (b) encompassing joints and; (c) Illustra- tion of the cascade inverse kinematics algorithm on a deformed sphere.

Experiments

Sphere Creation

We evaluate our method on the task of in-hand cube reorientation in both simulation and the real world.

Simulation Result

Sphere Creation

In-Hand Manipulation Sim-to-Real Results

We evaluate dexterous manipulation on LEAP Hand and Allegro Hand in real-world in-hand cube reorientation tasks.

LEAP Hand

Training with all four hands: 1.1 MEAN Reposes

Training on LEAP hand: 2.0 MEAN Reposes

Zero-shot: Training on all hands except the target hand: 0.9 MEAN Reposes

Joint control baseline: 0.6 MEAN Reposes

Allegro Hand

Training with all four hands: 2.1 MEAN Reposes

Training on Allegro hand: 2.1 MEAN Reposes

Zero-shot: Training on all hands except the target hand: 0.8 MEAN Reposes

Code

Please check the code below.

BibTeX

Please cite UHAS if it helps your research:

      @misc{casas2026crossembodiment,
        title={Cross-Embodiment Robot Manipulation via a Unified Hand Action Space}, 
        author={Luis Felipe Casas and Robert Teal and Keval Shah and Abhijit Tadepalli and Wanxin Jin and Yu Xiang},
        year={2026},
        eprint={2607.03570},
        archivePrefix={arXiv},
        primaryClass={cs.RO},
        url={https://arxiv.org/abs/2607.03570} 
      }

Contact

Send any comments or questions to Luis Felipe Casas:
Luis.CasasMurillo@utdallas.edu

QR code for the UHAS project webpage

Scan Project Page

Open the UHAS website on another device.

Acknowledgements

This work was supported in part by the National Science Foundation (NSF) under Grant Nos. 2346528 and 2520553, the NVIDIA Academic Grant Program Award, and a gift funding from XPeng.