Grasp Exploration for 3D Object Shape Representation Using Probabilistic Map

Abstract

Abstract. In this work, the representation of 3D object shape acquired from grasp exploration is presented. Electromagnetic motion tracking sensors are attached to the fingertips for object contour following, acquiring thus the 3D points to represent the object shape using a probabilistic volumetric map. The object-centric representation is adopted by finding the centroid through 3D moment invariants. The occupancy of each individual voxel in the map is assumed to be independent from the other voxels occupancy, and the posterior information is obtained by using the Bayes' rule, thus achieving the probability distribution on the occupations percentage for each voxel. The conversion of the probabilistic map in Cartesian space to the spherical coordinate system for visualization with more resolution in an egocentric view is proposed.