A survey and experimental study of neural network AUV control

Abstract

This paper presents a survey of neural network controllers for AUVs. A direct adaptive neural net controller incorporating integral action was designed for the heave motion of the ODIN underwater vehicle. The neural net controller is trained online by parallel recursive error prediction method and the critic equation. The influence of the various design parameters of the neural net controller was investigated by computer simulation and the controller was also experimentally tested using the ODIN. Results of the computer simulation and experiment are discussed in this paper.