GenDexHand: Generative Simulation for Dexterous Hands

TL;DR

GenDexHand is a novel generative pipeline that autonomously creates diverse, high-quality dexterous manipulation tasks and environments, enabling scalable training of complex robotic hand behaviors through a closed-loop refinement process and task decomposition.

Contribution

It introduces a closed-loop refinement process using vision-language feedback and task decomposition to generate feasible, diverse dexterous manipulation environments, addressing data scarcity in embodied intelligence.

Findings

Improved environment quality through VLM-based refinement.

Enhanced training efficiency via task decomposition.

Scalable generation of diverse dexterous manipulation tasks.

Abstract

Data scarcity remains a fundamental bottleneck for embodied intelligence. Existing approaches use large language models (LLMs) to automate gripper-based simulation generation, but they transfer poorly to dexterous manipulation, which demands more specialized environment design. Meanwhile, dexterous manipulation tasks are inherently more difficult due to their higher degrees of freedom. Massively generating feasible and trainable dexterous hand tasks remains an open challenge. To this end, we present GenDexHand, a generative simulation pipeline that autonomously produces diverse robotic tasks and environments for dexterous manipulation. GenDexHand introduces a closed-loop refinement process that adjusts object placements and scales based on vision-language model (VLM) feedback, substantially improving the average quality of generated environments. Each task is further decomposed into sub-tasks to enable sequential reinforcement learning, reducing training time and increasing success rates. Our work provides a viable path toward scalable training of diverse dexterous hand behaviors in embodied intelligence by offering a simulation-based solution to synthetic data generation. Our website: https://winniechen2002.github.io/GenDexHand/.