Model recovery of unknown objects from discrete tactile points

Abstract

This paper presents a novel method for robot to reconstruct unknown object models from discrete tactile point clouds. In model recovery process, 6D tactile data which contain point positions and corresponding normal vectors are firstly clustered. Then geometric feature descriptors are used to extract geometric feature vectors from tactile data. The object feature vectors are used in object shape classification and unknown object models are recovered. Simulation and experiment show that the present approach can be used to recognize unknown object shapes from sparse and noisy tactile point clouds and reconstruct object model accurately.