A neural-network-based controller for a single-link flexible manipulator: Comparison of FFNN and DRNN controllers

This paper employs two types of neural networks to control a single-link flexible arm. To train each network, we utilize a gradient-based approach with adaptive learning rate. We first apply the diagonal recurrent neural network (DRNN) to a single-link flexible arm, which is a challenging control problem, in order to investigate the ability of this type of recurrent neural network. We then apply a feed-forward neural network (FFNN) to this problem and perform some case studies for the purpose of performance comparisons of the two structures. Several simulations presented in this paper verify that the DRNN-based controller significantly improves the precision of the tip motion tracking, suppresses the tip deflections of the manipulator more effectively and simultaneously produces more appropriate control voltages.