Neural networks as robot arm manipulator controller

Abstract

Traditional controllers used in robot arm manipulators are complex, nonadaptive, and somewhat slow. Several researchers have developed approaches that use neural networks as controllers. This paper includes a theoretical discussion and design details for a robot arm manipulator controller using a neural network trained by feedback error learning, originally proposed by Kawato. The scheme and technique used in this research differ from the work published earlier (Kawato, Furukawa, and Suzuki 1987, Kawato, Isobe, Mayeda, and Suzuki 1988, and Lippman, 1987), although analyses and implementation presented here combine the best of them. The feedback learning scheme was implemented for robot arm manipulator with three degrees of freedom. The results of simulation were compared with the desired trajectory given.<>