FunctionalGrasp: Learning Functional Grasp for Robots via Semantic Hand-Object Representation
Yibiao Zhang, Jinglue Hang, Tianqiang Zhu, Xiangbo Lin, Rina Wu, Wanli Peng, Dongying Tian, Yi Sun
Abstract
Successful grasp is an important and long-standing issue for robots to interact with the real world. Most recent studies have devoted more attention to stable grasp rather than functional grasp, which cannot guarantee task-oriented postgrasp manipulation. To achieve human-like functional grasp, a semantic representation of functional hand-object interaction is introduced without labeling 3D hand poses, and a novel coarse-to-fine grasp generation network is designed to model this hand-object interaction. First, a coarse grasp is generated by combining the global hand pose and hand grasp type. Then, the fine pose will be optimized by guiding each finger to focus on the corresponding functional region of the object. Experimental results demonstrate the effectiveness of our method in achieving functional grasps for dexterous hands in the absence of high-DoF grasp poses annotation of the hand. The project website is: <uri xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">https://github.com/zhangyb1008/FunctionalGrasp</uri> .