DEVELOPMENT OF A SMALL CRUISING-TYPE AUV “MANTA-CERESIA” AND GUIDANCE SYSTEM CONSTRUCTED WITH NEURAL NETWORK

A small cruising-type testbed vehicle named “Manta-Ceresia ” is developed to validate performance of various control and guidance architectures for cruising-type Autonomous Underwater Vehicles (AUVs). It is so compact (13.8 kg in air) that one researcher can carry and handle the vehicle. Despite of smallness it is equipped with fundamental sensors and actuators required for prototype AUVs. Pairs of thrusters and elevators make it possible to swim in 3 dimensional space. The vehicle can swim along a wall of a square pool keeping constant distance automatically making use of 6 channels of range finder. Thus, the vehicle can continue swimming in a test tank as far as its energy remains. Taking advantage of these capabilities, adaptive controllers which require comparatively long swimming for adjustment can be examined in a small pool. Performance of a constant-altitude-controller system utilizing neural-network, which can accumulate experience by adaptation, is examined with the developed vehicle. Switching structure of the neural net-works is introduced to keep experience which is apt to be forgotten through additional learning and to represents mappings more precisely with same amount of learning calculation. Results of experiments with the developed testbed vehicle show that the introduced switching structure represents underwater terrain more precisely. Consequently, the controller is automatically adjusted to be applied for more complicated topography