Feedback error learning for control of a robot using SMENN

Abstract

The use of a new recurrent neural network (SMENN) employing feedback error learning for control of a robot is presented in this paper. The control system consisted of a feedback (PID) controller and two recurrent neural-network-based joint controllers. The network was trained using standard BP method as a learning algorithm. The effectiveness of the neural network was tested using different parameters of the robot. Diagonal neural network (DNN) was also employed as controllers of the robot in order to obtain comparisons with the proposed neural network. Moreover, the feasibility of the proposed neural controller (NC) is demonstrated through the simulation of the two-degrees-of-freedom SCARA type robot. Simulation results show the significant improvement of learning time and accuracy, which practically enables the use of NC in robotics applications.