Control of Underwater Autonomous Vehicles Using Neural Networks

Abstract

This paper describes a control method for low level control of autonomous underwater vehicles using a neural network approach. It presents an alternative to classical control methods currently used on the University of Idaho's miniature submarine and underwater crawler. The models of these vehicles are used in training an neural network that optimize the navigation control for the submarine and the track control of the underwater crawler. The effect of using neural networks for the control of such vehicles is its trainability to non-linear systems. Both vehicles operate in transverse conditions and intelligent control is paramount. The underwater conditions can change very rapidly from variables such as environmental aspects to the specific terrain the vehicle is transversing. Because of this environment these controllers should be adaptive to any situation that may arise and thus neural network control is well suited for robust control in this application