Event

Proposes cross-embodiment dexterous grasping framework MANGO-Grasp

1 reports · 1 sources

Heng Zhang et al. propose MANGO-Grasp, a cross-embodiment dexterous grasping framework that represents objects with geometry-oriented 3D Gaussian primitives, encodes hands as pretrained descriptors that capture both morphology and kinematics, and models hand-object contact with an anisotropic Mahalanobis field. The authors report that on the CMAP and MultiGripperGrasp benchmarks, the method outperforms the strongest seen-hand baseline in simulation by up to 8.24 percentage points; in zero-shot transfer to the unseen SharpaWave hand, it outperforms the strongest zero-shot baseline by up to 16.57 percentage points, with an 86% success rate in real-robot experiments. The code and supplementary materials are stated to be released upon publication of the paper.