GeoMatch++: Morphology Conditioned Geometry Matching for Multi-Embodiment Grasping

2024-12-25

AI Summary

Existing cross-morphology dexterous grasping methods exhibit poor generalization to unseen end-effectors and fail to model deep structural-geometric couplings between gripper morphology and object geometry. To address this, we propose a morphology-aware geometric matching framework that, for the first time, explicitly encodes gripper joint-link topology and kinematics as attention-guided geometric features, enabling zero-shot cross-morphology grasping policy transfer. Our method integrates an attention-driven morphology-conditioned matching network, multi-body dynamics-aware feature extraction, and cross-morphology geometric consistency constraint learning. Evaluated on three out-of-distribution end-effectors, it achieves an average grasp success rate improvement of 9.64% over state-of-the-art approaches. This work establishes a scalable, joint morphology-geometry modeling paradigm for general-purpose dexterous manipulation.

Technology Category

• Intelligent Robots: Manipulation
• Computer Vision: Multi-modal Vision
• Knowledge Representation and Reasoning: Geometric, Spatial, and Temporal Reasoning

Application Category

• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Search and Retrieval-Augmented AI: Multilingual and cross-lingual Web search

Abstract

Despite recent progress on multi-finger dexterous grasping, current methods focus on single grippers and unseen objects, and even the ones that explore cross-embodiment, often fail to generalize well to unseen end-effectors. This work addresses the problem of dexterous grasping generalization to unseen end-effectors via a unified policy that learns correlation between gripper morphology and object geometry. Robot morphology contains rich information representing how joints and links connect and move with respect to each other and thus, we leverage it through attention to learn better end-effector geometry features. Our experiments show an average of 9.64% increase in grasp success rate across 3 out-of-domain end-effectors compared to previous methods.

Problem

• Multi-fingered grasping
• Unseen manipulators
• Object shape recognition

Innovation

• GeoMatch++
• Attention Mechanism
• Unseen Hand Adaptation