Handling Unanticipated Events in Single and Multiple AUV Systems

An intelligent controller for AUVs must be able to handle unanticipated events as they occur. If not, it risks the mission and the vehicle. Unfortunately , this is a diicult, context-dependent task: the agent must rst detect that there has been an event, then diagnose it, assess its likely importance, and nally select an appropriate response and integrate that response into its other actions. Event handling is even more diicult in situations involving multiple cooperating AUVs, since now the agent must analyze the event in terms of its likely impact on the others and the group as a whole. In this paper, we discuss two versions of event handling in Orca, a schema-based AUV controller for ocean science and cooperative distributed problem solving missions. We then discuss how the process can be extended for multiagent event handling. During a mission, an intelligent autonomous underwater vehicle (AUV) controller will encounter unanticipated events. Sensors will return uncertain and unexpected information , the controller will have incomplete and uncertain knowledge about the world around it and about itself, and there will be unknown, unpredictable agents and processes interacting with it. Unless these unantic-ipated events are recognized and appropriately handled, they can jeopardize the AUV's mission or even the vehicle itself. By \unanticipated," we do not mean that an event is necessarily novel or even completely unexpected. Instead , it simply cannot be predicted well enough to plan for it. For example, \power failed," \collision," \mo-tion stopped," and so forth are all unanticipated events in the AUV domain; they are to some extent expected to occasionally occur, but predicting exactly when and how would be very diicult. Not all events that occur are unanticipated, of course, nor are all harmful. Anticipated events are easily handled|responses, if any are needed, can be determined when they are predicted| and beneecial unanticipated events will be a topic for future research. In this paper, we restrict our attention to potentially harmful unanticipated events. Handling unanticipated events is diicult. They are sometimes hard to detect, and it is even harder to decide what caused them, how important they are, and how to respond to them. If the AUV is part of a cooperative distributed problem solving (CDPS) system, the problem is compounded. The AUV controller must now be concerned about what the event means for the others, the group as a whole, and their shared mission. Multiagent event handling …