Development of motion control system for AUV using neural nets

A self-organizing neural-net-controller system is developed as an adaptive motion control system for AUVs (autonomous underwater vehicles). The system can generate autonomously an appropriate controller according to some evaluations of motion of the vehicle. To construct the system in practical environment, an organizing process of the controller is proposed, which is initiated by a premature fuzzy controller. This process enables the system to generate the controller with minimum prior knowledge of the effect of control actions. The system was applied to the longitudinal motion control of an AUV and succeeded in self-organization of the controller through free-swimming tank tests