A comparison of neural network controllers for a mobile robot with an on-board manipulator

Abstract

A systematic approach for modeling and motion control of a mobile vehicle with on-board arm is presented. Two neural network based controllers which feedback linearize the composite system after the incorporation of non-holonomic constraints are considered. The feedback linearization provides an inner loop that accounts for possible motion of the on-board arm. These neural network controllers exhibit learning-while functioning features instead of the traditional learning-then-control training approach. Therefore, the control action is immediate with no off-line-learning phase needed. The case of maintaining a desired course and speed while the on-board arm is allowed to move to its desired orientation is considered. The two neural network algorithms used in designing the controller are backpropagation with e-mod and Hebbian learning with e-mod. Computationally the Hebbian learning with e-mod outperforms the backpropagation with e-mod without any performance degradation. A computational comparison and simulation results are presented in order to justify the theoretical conclusion.