Neural network based control of underactuated manipulators

Abstract

We propose in this work a neural network based control method to position the passive joints of a mechanical manipulator. Manipulators with such types of joints are known in the literature as underactuated manipulators. A standard feedforward neural network (FNN) is trained to learn the dynamic behavior of the manipulator from random state space data generated off-Iine by a variable structure controller. After learning is completed, the FNN is utilized for real time control of the underactuated manipulator's unactuated joint. Simulation results are presented to demonstrate the validity of the proposed method.