LAUSR

A Mobile manipulator (MM) combines the ability to move on an uneven terrain and perform a desired task using the manipulator arm, like drilling, welding, pick and place, etc. However, due to slip, modeling and sensor errors during motion, the MM tends to deviate from the desired end effector path. The MM considered in this paper consists of a 10 DOF vehicle and a 4 DOF arm. As the system is redundant, the standard jacobian based pseudo inverse methods for finding the inverse kinematics solution cannot be used in real time and are very complex. Therefore, a learning based method using a two stage KSOM network has been used to learn the inverse kinematics solution and also track the end effector during motion. In the first stage the KSOM performs the redundancy resolution of the system and in the second stage it corrects the error during the actual motion. The forward kinematics model of the mobile manipulator is used to train the KSOM network, with the manipulability measure of the arm, joint angles of the system and the wheel terrain contact as constraints. Once the network is trained for a particular terrain, the corresponding joint angles and wheel velocities for tracking a desired end effector trajectory can be found in real time. Simulation and experimental results for tracking different end effector trajectories on an uneven terrain, proves the efficiency of the proposed method. The experimental results prove that the proposed method could track the end effector trajectory with an average error of less than 2 cm.