Workspace-Based Precision Grasp Pose Generator for Multi-fingered Robotic Hands

A valid precision grasp pose can be seen as a combination of relative pose between the palm and the object, contact points of fingertips and joint angles for each fingertip to reach the contact point. In this paper, point clouds of fingertip workspaces under the palm coordinate system are utilized to bind all the above mentioned information within one representation. To speed up grasp pose generation, point clouds of workspaces are preprocessed based on precision grasp configurations. Finally, a sampling based method is proposed to generate precision grasp poses for multi-fingered robotic hands. Experiments have been done in both simulation and real world to verify that our methods can be applied for precision grasp pose generation on robotic hands with various structures.