Adaptive robust robot control using BP-SMENs

Abstract

This paper presents the development of a new adaptive recurrent neural network for the control of a nonlinear system represented by a two-link SCARA type planar robot manipulator. The standard backpropagation algorithm is used to adjust the weights of the networks. The proposed control system consists of an inverse neural model of robot (INNM), an INNM-based neural controller, a robust controller, a conventional PI controller, and a second order linear filter. To evaluate the performance of the proposed control scheme and neural network, a simulated SCARA type robot was studied and the results showed how well the proposed controller can minimise the error between an actual and desired end-effector trajectory. From simulation examples, the robot trajectory tracking showed superior performance that is very attractive for real-time implementation and application in complex industrial tasks. For comparison, the standard computed torque method is employed for controlling the robot.