Range-map based grasp planning for a bin of cylindrical parts

Abstract

This thesis describes a system which determines the best grasp vector to acquire randomly oriented cylindrical parts piled in a bin using a parallel jaw gripper. Specifically, we address the problems of the range map preprocessing and grasp planning, assuming that the range map is available by a ranging technique. In the preprocessing module, we identify the individual parts from a range map using the characteristics of local surface curvatures and search non-occluded parts from those parts using the information of each part``s boundary, i.e., concavity checking. In grasp planning module, we search for some space around each part for gripper placement to find graspable parts and sort all graspable parts by using stability function to determine the best grasp vector. In addition to the algorithms for the two modules, we report on the results of experiments using synthetic range maps.