A Neural Network Based Self-learning Control System for Underwater Robots

Abstract

This paper describes a neural network based self-learning control system developed especially for autonomous underwater robots. Artificial neural networks can be effective tools for handling difficulties such as nonlinear dynamics and unpredictable disturbances which are often involved in the control problems of underwater robots. Inspired by growing-up processes of biological creatures, a self-learning architecture of neural net controllers has been established by the authors. The principles of the architecture called “SONCS: Self-Organizing Neural-net-Controller System” is presented and the system's feasibility is examined through several real-world applications which include constant depth and altitude swimming of a cruizing type autonomous underwater robot “PTEROA”, and vibration control of a multi-degree of freedom structure with a damping controllable dynamic damper. It is shown that the neural net controllers can be appropriately adjusted by the proposed self-learning procedures and that the SONCS can be applied to a wide variety of control problems with just easy modifications. New ideas and future perspectives, which include a newly developed quick adaptation methods called “Imaginary Training”, are discussed to make the SONCS system more attractive solution for control engineering of actual robotic systems.