Learned anticipation strategy for speed control in an AUV fleet

Researchers at the Laboratory of Artificial Intelligence and Robotics (LAIR) and the Center for Intelligent Systems Research (CISR) from the University of Idaho (UI) have developed a message anticipation module for use by members of a fleet of autonomous underwater vehicles (AUV). The test scenario is a magnetic signature assessment (MSA) task, in which a fleet of five AUVs must simultaneously pass under a moving Target Ship (TS) at a predetermined location. During the task the TS informs the AUVs regarding its progress, allowing the AUVs to meet the TS at that measurement point despite variations in the TS´s velocity. However, the underwater acoustic modems used by the actual AUVs are both low bandwidth and noisy. Thus, messages from the TS may be infrequent or erroneous. The goal of the anticipation module is to anticipate the TS´s messages and, when necessary, use the anticipated message to fill in gaps left by dropped or erroneous messages. Successful anticipation depends on an agent having a good knowledge of its environment and mission. Research has shown that anticipation of words and sentences is central to human communication and language understanding. An agent that utilizes similar anticipation methods and is capable of using artificial intelligence techniques to generate, utilize, and adapt its message anticipation module with a more accurate one, would represent a significant advance in autonomous agents. Five different anticipation models were created, 4 of them are based on a neural network model and one of them is based on a fuzzy logic controller model. In order to test effectiveness, robustness, and adaptability of the models used in the anticipation module, multiple tests were conducted in which the behavior of the target ship and the gap between messages were varied. All of the tested anticipation models were able to significantly reduce the error in the meeting point when there was a gap between messages.