Implementation of machine learning algorism to autonomous surface vehicle for tracking and navigating AUV

On the construction of Kanda port in Fukuoka prefecture, many harmful chemical bombs have been discovered beneath the sea bottom and they are needed to be dug up carefully and quickly as possible. So our group is trying to develop a new sub-bottom interferometric synthetic aperture imaging sonar (sub-bottom interferometric SAS) system to recognize chemical bombs as centimeters-resolution 3D-sub-bottom acoustic image. In this R&D study, it is the reasonable methodology to use an autonomous underwater vehicle (AUV) which can survey the seafloor with sub-bottom interferometric SAS transmitter and receiver at a constant height. To accomplish this R&D goal, positioning AUV accurately is needed, so we are trying to develop the technique which minimizes the error of positioning, using autonomous surface vehicle (ASV) which tracks AUV and surveys its absolute position by super short-baseline (SSBL) method. In development of tracking ASV, it is important to develop the controlling algorism which orders ASV to steer stably and control adequately its velocity according to the result of SSBL positioning of the AUV. Based on machine learning method, we are trying to develop an algorism which infers appropriate control of ASV from precious controlling log. Implementation of this algorism will improve the precision of underwater positioning. This paper reports the development status of our ASV and controlling algorism.