Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
This paper introduces GraspGraphNet, a novel framework for dexterous grasp generation that leverages a topology-aware approach by representing robot hands as URDF-derived kinematic graphs. The method integrates hierarchical object surface encoding, differentiable forward kinematics, and dynamic world-edge message passing to directly generate executable palm poses and joint configurations, achieving an impressive average success rate of 83.48% on a 40-object benchmark. Notably, the model demonstrates robustness to variations in hand topology, achieving a 72.70% success rate on finger-removal variants without retraining, highlighting the efficacy of graph-structured representations in robotic grasping tasks.
Achieving an 83.48% success rate in dexterous grasp generation across diverse robot hands without the need for post-processing optimization is a game-changer for robotic manipulation.
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