GraspGraphNet: Graph-Structured Multi-Embodiment Dexterous Grasp Generation

2026-07-12

AI Summary

This work addresses the challenge of grasp generation for heterogeneous dexterous hands, which arises from differences in kinematic topology, actuation dimensionality, and command spaces. The authors propose a unified graph-based modeling approach that represents a robotic hand as a kinematic graph derived from its URDF, integrating hierarchical object surface encoding, differentiable forward kinematics, and dynamic graph message passing to directly generate feasible grasps in palm pose and joint space. Innovatively leveraging conditional flow matching, the method circumvents inverse kinematics, post-hoc optimization, and retargeting, thereby enabling cross-hand generalization. Evaluated on Barrett, Allegro, and Shadow hands, the approach achieves an average success rate of 83.48% with a single inference time of 40 ms, and without retraining, it attains a 72.70% success rate even on finger-joint–missing variants.

Abstract

Dexterous grasp generation across robot hands is challenging because hands differ in kinematic topology, actuation dimensions, and native command spaces. We introduce GraspGraphNet, a topology-aware grasp generation framework that represents each hand as a URDF-derived kinematic graph and directly generates executable palm poses and joint configurations. GraspGraphNet combines hierarchical object surface encoding, differentiable forward kinematics, and dynamic world-edge message passing to model evolving robot-object interactions. It applies conditional flow matching directly in executable palm-pose and joint-state space, avoiding post-processing optimization, inverse kinematics, and retargeting. Using a shared model trained on Barrett Hand, Allegro Hand, and Shadow Hand, GraspGraphNet achieves an average success rate of 83.48% with 40ms inference time per grasp on a 40-object benchmark. Without retraining, the same model achieves 72.70% success on controlled finger-removal variants, demonstrating robustness to hand-topology variations. These results suggest that graph-structured hand representations can effectively support dexterous grasp generation across robot hands with different kinematic structures. Project: https://lysees.github.io/graspgraphnet-page

Problem

• dexterous grasp generation
• multi-embodiment
• kinematic topology
• robot hands
• grasp transfer

Innovation

• graph-structured representation
• dexterous grasp generation
• conditional flow matching
• kinematic topology generalization
• differentiable forward kinematics