Learning framework for inverse kinematics of a highly redundant mobile manipulator

Abstract

A learning framework is proposed to solve the inverse kinematic problems of a highly redundant mobile manipulator designed to traverse on rough terrains. The problem is not trivial to solve and there does not exist a closed form solution. The learning framework is designed based on the Kohonen Self Organizing Map (KSOM) to establish the mapping between the task-space and joint-space while resolving redundancy exist in the system. The standard KSOM learning architecture is modified to ensure proper coordination between the mobile base and arm satisfying multiple constraints such as wheels of the mobile robot always remain in contact with the terrain and maximize the manipulability for the robot arm while solving inverse kinematics. The network is trained for 14 DoF mobile manipulator traverse on uneven terrain. This method can be extended to other types of robot, such as high degrees of the manipulator. To validate the proposed network architecture several simulations have been performed and experimented on a mobile manipulator considering robot traverse on different types of terrains. The results show the effectiveness of the proposed framework.