CAPABILITY REPRESENTATIONS OF MOBILE MANIPULATIONS AMIDST OBSTACLES

Abstract

Motion planning for a mobile manipulator becomes complex due to the increase in degrees of freedom and different behavioral characteristics of the base robot and the manipulator. Proper coordination between the mobile base and the manipulator can be achieved utilizing capability representations such as reachability maps and inverse reachability maps. The aim of this thesis was to propose methods to calculate accurate capability representations amidst environment obstacles and evaluate the impact of obstacles on the capability representations. This thesis proposed and evaluated several machine learning models (ReachNet and Inv-Reach Net) to calculate capability representations amidst obstacles. Results indicated that an obstacle’s region of influence on the capability representations extends far beyond the physical boundaries of the obstacle and the proposed models calculated capability representations faster with minimal loss in accuracy even when there were multiple obstacles. The proposed models could be used for mobile manipulator base placement.