Discrete-time parallel robot motion control using adaptive neuro-fuzzy inference system based on improved subtractive clustering

Abstract

This paper addresses a high precision discrete-time model-free PID adaptive neuro-fuzzy logic motion controller in case the physical models that describe a robot are not known. The advantage of this kind controller is that it uses an improved subtractive clustering technique to obtain the structure of the system model in order to ensure the high accuracy of the intelligent control. Moreover, the information used is directly from the nonlinear system response without the knowledge of the robot physical parameters and complex models. Furthermore, the controller is designed in discrete-time domain for allowing its implementation. To conceive this kind intelligent control, first adaptive neuro-fuzzy inference system with an improved subtractive clustering computing is used to accomplish the integration of information of joint angular displacement and velocity for torque identification. The learning datasets are generated by using a discrete-time PID feedback control, which desired position is sufficiently rich to ensure that the learning system would include most characteristics of the nonlinear dynamics of the system. Then a discrete-time fuzzy feed forward control, combined with a conventional PID and the adaptive neuro-fuzzy inference system, is designed for the mechanism motion control. Simulation results from numerical experiment on a 4-bar planar parallel mechanism show the proposed controller can reduce joint position and velocity tracking errors with higher accuracy and higher reliability than a traditional PID controller and a computed torque controller.