REGRIND: A Minimalist Retargeting-Guided RL Recipe for Dexterous Manipulation from Single Human Demonstrations

Forum topic · 小凯 · 2026-07-15

Summary

This post introduces REGRIND, a minimalist retargeting-guided reinforcement learning pipeline (arXiv:2607.11874) that learns dexterous manipulation policies from a single human demonstration. Building on the recipe popularized in humanoid whole-body control, the authors retarget human hand-object motion to robot references that preserve hand-object spatial and contact relationships, train a residual RL policy in simulation to track object-centric keypoints along the reference, and transfer the policy zero-shot to hardware via careful system identification. The resulting policies produce fluid, human-like behaviors on two different multi-finger hands, accomplishing contact-rich tool-use tasks such as operating scissors and turning a screwdriver. Systematic hardware experiments identify key factors affecting sim-to-real transfer in dexterous manipulation, offering practical guidance for retargeting-based learning in contact-rich scenarios.

Paper Overview

Field: Robotics (RO)
Authors: Yunhai Feng, Natalie Leung, Jiaxuan Wang, Lujie Yang, Haozhi Qi
Published: 2026-07-13
arXiv: 2607.11874

Abstract (Translated Summary)

Recent work in humanoid whole-body control has found success with a simple recipe: retarget human motion to robot kinematic references, then train policies via reinforcement learning (RL) to track them. But how does this recipe transfer to dexterous manipulation? The answer is not obvious, as manipulation involves complex, contact-rich dynamics and requires delicate regulation of contact modes and forces.

The authors present REGRIND, a minimalist retargeting-guided RL pipeline that learns dexterous manipulation policies from a single human demonstration:

1. Retargeting: Human hand-object motion is retargeted to a robot reference that preserves hand-object spatial and contact relationships. 2. Residual RL: A residual RL policy is trained in simulation to track object-centric keypoints along that reference. 3. Sim-to-real transfer: The resulting policy is transferred zero-shot to hardware via careful system identification.

The resulting policies produce fluid, human-like behaviors on two different multi-finger hands, completing contact-rich tool-use tasks including operating scissors and turning a screwdriver. Through systematic hardware experiments, the authors identify and analyze the key factors affecting sim-to-real transfer in dexterous manipulation, providing practical guidance for retargeting-based learning in contact-rich scenarios.

Original Abstract (Excerpt)

> Recent work in humanoid whole-body control has found success with a simple recipe: retarget human motion to robot kinematic references, then train policies via reinforcement learning (RL) to track them. But how does this recipe transfer to dexterous manipulation? The answer is not obvious, as manipulation involves complex, contact-rich dynamics and requires delicate regulation of contact modes and forces. We present REGRIND, a minimalist retargeting-guided RL pipeline that learns dexterous manipulation policies from a single human demonstration. REGRIND retargets human hand-object motion to a robot reference that preserves hand-object spatial and contact relationships, trains a residual RL policy in simulation to track object-centric keypoints along that reference, and transfers the result...

---

*Auto-collected on 2026-07-15.*

This page is an English static mirror generated for search and AI citation. It may be a full translation or structured summary of the Chinese original. Canonical interactive discussion lives on the Chinese page: https://zhichai.net/topic/178395144