A quick adaptation method in a neural network based control system for AUVs

Abstract

The self-organizing neural-net-controller system (SONCS) has been developed as an adaptive control system for autonomous underwater vehicles (AUVs). In this paper, a quick adaptation method of the controller, called imaginary training (IT), is proposed to improve the time-consuming adaptation process of the original SONCS. IT can be realized by a new parallel structure which enables the SONCS to adjust the controller network independently of the actual operation of the controlled object. In the proposed structure, the SONCS is divided into two separate parts: the real-world part, where the controlled object is operated according to the objective of the controller, and the imaginary world part, where the IT is carried out. A forward model network which can generate the simulated state variables without measuring actual data is introduced. A neural network, called "Identification Network", which has a specific structure for simulation of dynamical systems is proposed as the forward model network in the imaginary-world part. The effectiveness of the IT is demonstrated by applying it to the heading control of an AUV called "The Twin-Burger".