Learning Dexterous Manipulation Using Contact Wrench Guidance From Human Demonstration

2026-06-22

AI Summary

This work addresses the challenge of effectively transferring human demonstrations to robot policies in dexterous manipulation, particularly for contact-rich and long-horizon tasks. The authors propose CHORD, a novel framework that introduces an object-centric contact wrench space as an intermediate representation for cross-modal action alignment. By mapping both human and robot actions into the forces and torques applied on the object, CHORD enables meaningful action similarity measurement and guides reinforcement learning. The approach generalizes from hand-centric or third-person video demonstrations to full-body robotic manipulation and is evaluated on a large-scale simulation benchmark comprising 4,739 bimanual manipulation tasks. Experiments demonstrate an average success rate of 82.12% across 1,831 tasks, 90.77% generalization performance to full-body manipulation, and successful open-loop and closed-loop deployment in real-world settings.

Abstract

Dexterous robot manipulation can benefit from the abundance of human demonstrations, but transferring such demonstrations to robot policies remains challenging. We present Contact Wrench Guidance from Human Demonstration in Robotic Dexterous Manipulation (CHORD), a framework for long-horizon manipulation of rigid and articulated objects with reinforcement learning. The key idea is object-centric contact wrench space guidance: we represent human and robot motions by the forces and torques they can induce on the object, enabling similarity to be measured by the induced instantaneous motions. This guidance makes reinforcement learning more scalable for contact-rich dexterous manipulation. We further introduce a large-scale simulation benchmark with 4,739 bimanual dexterous manipulation tasks, constructed from motion-capture datasets and reconstructed in-house videos. Evaluated on 1,831 benchmark tasks, CHORD achieves an average success rate of 82.12%, demonstrating strong scalability. CHORD also generalizes to whole-body manipulation from hand-only and third-person demonstrations, achieving a 90.77% success rate, and the learned policies transfer to the real world in both open-loop and closed-loop settings.

Problem

• dexterous manipulation
• human demonstration
• contact-rich
• reinforcement learning
• policy transfer

Innovation

• contact wrench guidance
• dexterous manipulation
• human demonstration
• reinforcement learning
• simulation benchmark