A 3D-grasp synthesis algorithm to grasp unknown objects based on graspable boundary and convex segments.

This paper presents a three dimensional (3D) grasp synthesis algorithm to achieve distinguished grasps supporting both stability and human-like grasping. The algorithm, which is based on the concepts of a graspable boundary and convex segments, was developed to enable a two-fingered gripper to grasp any unknown object, regardless of its shape, texture, or concavity, given a single 3D image data from depth sensors. The proposed algorithm provides ways to grasp any object using boundary, envelope, and functional grasps. The algorithm is based on identifying graspable segments, analyzing them geometrically, and incorporating the memory of grasping experience. Unlike most grasp synthesis research that focuses on complete 3D contours, our algorithm concentrates only on the graspable boundary and convex segments and thereby achieves stable grasps with less computational complexity. The experimental results show that the proposed algorithm provides distinguished and stable grasps for various objects in various environments, and is suitable for robots to grasp the objects successfully.