Self-tuning of a Neuro-Adaptive PID Controller for a SCARA Robot Based on Neural Network

Abstract

In this paper a MIMO (Multiple-Input-Multiple-Output) adaptive neural PID (AN-PID) controller that can be applied to a nonlinear dynamics is proposed, and its use is shown in the control of a SCARA robot for two degrees of freedom. The AN-PID controller, including a neural network of the dynamic perceptron type, is designed. The proposed controller uses a RBF network to identify the model and back propagates the control error to the AN-PID controller, unlike other controllers, that use direct methods to back propagate such error. With these properties, an AN-PID controller corrects the tracking errors due to the uncertainties and variations in the robot arm dynamics. It is robust and with adaptive capacity in order to achieve a suitable control performance. Experimental results on the SCARA robot were obtained to illustrate the effectiveness of the proposed control strategy, including comparison with a classical PID. By using Lyapunov's discrete-time theory, it was demonstrated that the control error is semi-global uniformly ultimate bounded (SGUUB).