Representation methods for recognizing and grasping object of robots

Recognizing and grasping object of robots are two crucial issues for robots accomplishing complicated tasks in unstructured environments, where object representation is a key premise problem. In this paper, we study the representation methods for (1) object recognition, SHOT descriptor is further described by using Tsallis entropy, which greatly reduces the storage and the computation time; (2) grasp part selection, a hierarchical grid representation method is proposed, which retrieves the abrupt concave area layer by layer for segmentation and builds layer minimum bounding cubes referring to the principal axis of point set for orientation changeable representation. Simulations and experiments have verified the effectiveness of the proposed methods.