Robots Learn Dexterous Manipulation from Human Demos with Contact Wrench Guidance

Xinghao Zhu, Zixi Liu, Shalin Jain, Chenran Li, Milad Noori, Huihua Zhao, John Welsh, Michael Andres Lin, Wei Liu, Tingwu Wang, Xingye Da, Zhengyi Luo, Vishal Kulkarni, Naema Bhatti, Yuke Zhu, Linxi Fan, Bowen Wen, Danfei Xu, Soha Pouya, Yan Chang· July 2, 2026 View original

Summary

This paper introduces CHORD, a framework that enables robots to learn complex, long-horizon dexterous manipulation tasks from human demonstrations using reinforcement learning. It employs object-centric contact wrench space guidance to measure motion similarity and scales effectively across a large simulation benchmark, also transferring to real-world applications.

Teaching robots complex, dexterous manipulation skills, especially those involving contact with objects, remains a significant challenge. While human demonstrations offer a rich source of data, effectively transferring these skills to robot policies is difficult. Researchers have developed a new framework called CHORD (Contact Wrench Guidance from Human Demonstration in Robotic Dexterous Manipulation). CHORD addresses this by representing both human and robot motions through the forces and torques they exert on an object, allowing for a more precise measurement of motion similarity. This "contact wrench space guidance" makes reinforcement learning more scalable for intricate, contact-rich tasks. The framework was evaluated on a massive simulation benchmark of nearly 5,000 bimanual manipulation tasks, achieving high success rates. CHORD also demonstrated generalization to whole-body manipulation and successful transfer of learned policies to real-world robots, both in open-loop and closed-loop control.

Why it matters

This advancement significantly improves the ability of robots to perform complex, human-like manipulation, opening doors for automation in industries requiring fine motor skills and intricate object handling. Professionals can anticipate more versatile and capable robotic systems.

How to implement this in your domain

  1. 1Explore integrating advanced dexterous manipulation capabilities into existing robotic systems for assembly or handling tasks.
  2. 2Investigate the potential for using human demonstrations to rapidly train robots for new, complex operational procedures.
  3. 3Pilot CHORD-like frameworks in simulation environments to assess their applicability to specific manufacturing or logistics challenges.
  4. 4Collaborate with robotics researchers to understand the practical implications and deployment strategies for contact-rich manipulation.

Who benefits

ManufacturingLogisticsHealthcareAutomotiveConsumer Electronics

Key takeaways

  • CHORD enables robots to learn dexterous manipulation from human demonstrations.
  • It uses object-centric contact wrench guidance for scalable reinforcement learning.
  • The framework achieves high success rates in complex bimanual tasks in simulation.
  • Learned policies successfully transfer to real-world robot applications.

Original post by Xinghao Zhu, Zixi Liu, Shalin Jain, Chenran Li, Milad Noori, Huihua Zhao, John Welsh, Michael Andres Lin, Wei Liu, Tingwu Wang, Xingye Da, Zhengyi Luo, Vishal Kulkarni, Naema Bhatti, Yuke Zhu, Linxi Fan, Bowen Wen, Danfei Xu, Soha Pouya, Yan Chang

"arXiv:2607.00033v1 Announce Type: cross 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 Roboti…"

View on X

Originally posted by Xinghao Zhu, Zixi Liu, Shalin Jain, Chenran Li, Milad Noori, Huihua Zhao, John Welsh, Michael Andres Lin, Wei Liu, Tingwu Wang, Xingye Da, Zhengyi Luo, Vishal Kulkarni, Naema Bhatti, Yuke Zhu, Linxi Fan, Bowen Wen, Danfei Xu, Soha Pouya, Yan Chang on X · view source

Want to go deeper?

Turn these trends into skills with Learnijoy's hands-on AI & tech courses.

Explore courses

More in AI Research

AI ResearchAI Engineering & DevTools

Human Feedback Guides Generative Meta-Learning for Robust Generalization.

This paper introduces Generative Meta-Learning with Human Feedback (GMHF), a framework that uses expert intuition to guide data synthesis and bridge the domain gap for machine learning models. GMHF employs a Conditional Neural ODE as a generative digital twin and an RL agent to refine latent physical parameters based on feedback, significantly reducing deployment loss and improving generalization under distribution shifts.

Midhun Parakkal Unni, Samuel KaskiJul 2, 2026
AI ResearchAI Engineering & DevTools

Valdi: Value Diffusion World Models for MPC

Valdi introduces Value Diffusion World Models, combining end-to-end online training for Model Predictive Control (MPC) with a latent diffusion dynamics model. Preliminary experiments show that Valdi, using a single diffusion step, matches deterministic MLP baselines in the CarRacing environment, highlighting a trade-off between predictive multimodality and control performance.

Christopher Lindenberg, Kashyap ChittaJul 2, 2026
AI Engineering & DevToolsAI Research

Task-Aware LLM Quantization Improves Efficiency and Performance.

This paper introduces TASA (Task-Aware Sensitivity Analysis), a two-level framework for mixed-precision quantization of large language models (LLMs) that optimizes calibration data composition and bit allocation. TASA addresses the "Perplexity Illusion" and the "Alignment-Diversity Tradeoff," enabling 3.5-bit models to match or surpass 4-bit baselines by jointly considering perplexity and reasoning-oriented sensitivity.

Fei Wang, Chao Xue, Taoran Liu, Li Shen, Ye Liu, ChangXing DingJul 2, 2026