Experimental Investigation of an Adaptive Neuro-fuzzy Control Scheme for Industrial Robots

This paper presents the application of an adaptive fuzzy logic controller with feed-forward component (AFLCF) to the Selective Compliance Assembly Robot Arm (SCARA Robot). The feed forward torque component is computed on-line using an artificial neural network (ANN) which has been trained off-line. This feed-forward component is designed to deliver the ideal torque component to the robot derivers. The feedback fuzzy logic control (FLC) component is made to keep the stability of the closed loop system. As the FLC is dependent in its rule base, here, a compact rule base is used. It consists of only four rules per each degree of freedom (DOF). The FLC ensures closed loop stability in the sense of Lyapunov and is valid for second order nonlinear systems. Furthermore, adaptability of the FLC has been achieved to enhance the tracking performance. The theoretical background of this control algorithm has been published in[1]. Using SCARA robot as the testing platform, here, experimental results are presented for the following five controllers: the conventional PD controller, PD controller tuned by fuzzy system (PDT), the FLC, Adaptive FLC (AFLC), and finally the AFLCF. The controllers are tested experimentally at the same initial conditions to make fair comparison between their performances. Results show that the investigated AFLCF has outperformed the other controllers.