SONCS: Self-organizing neural-net-controller system for autonomous underwater robots

Abstract

The self-organizing neural-net-controller system (SONCS) is introduced as a neural network based adaptive control system. The system includes a multilayered neural network called a forward model network which represents the dynamics of the controlled object. The basic idea is to adapt the controller with the evaluation and adaptation mechanism for estimating the object motion with the forward model. A multilayered neural network which has recurrent connections from the hidden layer to the input layer is proposed for the forward model. Characteristics of this forward model are investigated using a nonlinear system as a modeled object. The proposed network is available for estimation over a wide range of frequency. The SONCS was applied to the control problem of a small underwater robot, and its performance was examined through free-swimming tank tests.< <ETX xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">&gt;</ETX>