Advanced KSOM based Redundancy Resolution of a Mobile Manipulator System for Motion on an Uneven Terrain

Abstract

In this paper, we propose an advanced strategy for path following by a redundant mobile manipulator (Rover) using Kohonen self-Organizing Map(KSOM) based learning architecture. The rover consists of 10 DOF mobile platform with a 4 DOF manipulator mounted on top of it. The 14 DOF system is redundant and does not have a closed form inverse kinematics solution. In addition to the redundancy resolution, as the rover moves on uneven terrain the wheel and ground contact has to be ensured. The KSOM network is first trained using forward kinematics model of the rover manipulator system, with manipulability measure and joint angles of the manipulator serving as constraints. As compared to earlier KSOM methods an adaptive multistep correction is used in the learning loop. Simulation results of the end effector tracking different trajectories on various 3D terrain profiles is presented. The method shows superior performance than previous strategies in terms of accuracy achieved and reduced program execution time.