Learning Dexterous Manipulation with Quantized Hand State

2025-09-22

AI Summary

In dexterous robotic manipulation, high-dimensional hand actions dominate the coupled action space, degrading arm control performance. To address this, we propose a state-quantization-driven arm-hand decoupling learning method. Our core contribution is a discrete quantization encoding of hand states combined with continuous relaxation, which preserves fine-grained hand manipulation capability while significantly reducing action representation dimensionality. This enables construction of a compact joint action space tailored for diffusion-based policies, facilitating synergistic modeling and decoupled optimization of hand and arm actions. Within a data-driven framework, the method enhances policy learning stability and convergence efficiency. Experiments on complex dexterous manipulation tasks demonstrate superior generalization, balanced arm-hand control, and higher task success rates compared to baseline approaches.

Technology Category

• Intelligent Robots: Manipulation
• Search and Optimization: Learning to Search
• Machine Learning: Learning with Manifolds

Application Category

• Economics, Online Markets and Human Computation: LLM based quality controls for crowd work
• Responsible Web: Machine-in-the-loop, human agency and autonomy
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

Dexterous robotic hands enable robots to perform complex manipulations that require fine-grained control and adaptability. Achieving such manipulation is challenging because the high degrees of freedom tightly couple hand and arm motions, making learning and control difficult. Successful dexterous manipulation relies not only on precise hand motions, but also on accurate spatial positioning of the arm and coordinated arm-hand dynamics. However, most existing visuomotor policies represent arm and hand actions in a single combined space, which often causes high-dimensional hand actions to dominate the coupled action space and compromise arm control. To address this, we propose DQ-RISE, which quantizes hand states to simplify hand motion prediction while preserving essential patterns, and applies a continuous relaxation that allows arm actions to diffuse jointly with these compact hand states. This design enables the policy to learn arm-hand coordination from data while preventing hand actions from overwhelming the action space. Experiments show that DQ-RISE achieves more balanced and efficient learning, paving the way toward structured and generalizable dexterous manipulation. Project website: http://rise-policy.github.io/DQ-RISE/

Problem

• High-dimensional hand actions dominate coupled action space in dexterous manipulation
• Existing visuomotor policies compromise arm control due to combined action representation
• Learning arm-hand coordination is difficult with tightly coupled high-DOF motions

Innovation

• Quantizes hand states to simplify motion prediction
• Applies continuous relaxation for joint arm-hand diffusion
• Prevents hand actions from overwhelming the action space