Contact2Grasp: 3D Grasp Synthesis via Hand-Object Contact Constraint

• DOI: 10.24963/ijcai.2023/117
• Publication Date: 2023-08-11T08:31:30Z
• Authors: Haoming Li; Xinzhuo Lin; Yang Zhou; Xiang Li; Yuchi Huo; Jiming Chen; Qi Ye
• Keywords: Generality; Representation; Grippers

3D grasp synthesis generates grasping poses given an input object. Existing works tackle the problem by learning a direct mapping from objects to distributions of poses. However, because physical contact is sensitive small changes in pose, high-nonlinear between object representation valid considerably non-smooth, leading poor generation efficiency and restricted generality. To challenge, we introduce intermediate variable for areas constrain generation; other words, factorize into two sequential stages assuming that are fully constrained maps: 1) first learn map generate potential maps grasps; 2) then Further, propose penetration-aware optimization with generated contacts as consistency constraint refinement. Extensive validations on public datasets show our method outperforms state-of-the-art methods regarding various metrics.