Abstract
The underwater is fascinating fields for stand-alone and autonomous robots. They can enjoy swimming in the 3-dimensional world free from the restriction of gravity force. It is not easy, however, to construct reliable control systems for them which should be operated in the unknown and ever changing environment.
A newly developed “Self-Organizing Neural-Net-Controller System (SONCS)” is installed in the robot's brain. It can generate by itself an appropriate controller for swimming without assistance of superviser. The procedure of generation involves learning robot's dynamics associated by a premature Fuzzy controller, evaluation of swimming performance, and modification of controller to pursue a specified mission. The advantage of SONCS is demonstrated through free swimming tests in a pool. SONCS succeeds in constructing a pretty good controller to maneuver the PTEROA robot keeping constant depth and pitch as a result of several times of adaptation.
