In order to maneuver submerged vehicles (whether manned or unmanned) requires the use of many sensors and effectors. However, large numbers of sensors and/or effectors present a significant problem for human operators. The information overload that can take place in such instances presents a formidable obstacle. However, the use of expert systems can encapsulate much of the high level knowledge required to execute manuevers correctly based on sensor inputs. Reticular Systems proposes the development of an expert system capable of manuevering an underwater vehicle with little or no human interaction. To accomplish this we will leverage our previous Phase I efforts in which we developed a sophisticated simulation and test environment (STE) capable of interacting with an expert system to simulate the environment experienced by a vehicle. In addition, the expert system must respond in a timely manner. Therefore, Reticular Systems will utilize its experience in developing real-time AI techniques for expert systems to estimate predictable run-times and extend the temporal reasoning functionality. In addition, we propose investigating the potential of other AI techniques (neural networks, genetic algorithms, fuzzy logic) that will add flexibility and adaptability to the expert system so that the same rule base could be utilized over a range of different vehicles.
