Autonomous agents, whether humans or computer systems, must cope with the fact that the world is unpredictable. This has a variety of causes, most of which can be traced to uncertainty or incomplete information about the environment, unknown processes operating in the world, other agents' actions, or the actions of the agent itself. Unpredictability manifests itself in the form of unanticipated events that occur during problem solving: unexpected states or action outcomes, including failures and unanticipated interactions with others. Unless an agent can recognize and handle these events, it will be incapable of prolonged useful activity in the real world. Its plans will fail beyond recovery, and it will fail to seize unexpected opportunities.
Handling unanticipated events is difficult. Meaningful events are often difficult to pick out from the background of changes taking place in the world and the agent; even when detected, it is often hard to adequately diagnose the event's cause and select an appropriate response for the current problem-solving situation. The problem is compounded when an agent is part of a larger problem-solving ensemble, as in the case of a group of agents participating in cooperative distributed problem solving (CDPS). Here, agents must disentangle those events affecting only themselves from those affecting the broader group. In addition, at all points during the process of handling an event, an agent must keep in mind its role in the overall group and the fact that it may need to coordinate its information or responses with other members.
In this paper, we discuss event handling during multiagent cooperative problem solving. Our domain is cooperating autonomous underwater vehicles (AUVs) [Turner et al., 1991]; however, much of what we discuss should be applicable to any group of autonomous agents collaborating to solve problems in the real world. The work reported is largely a part of the ORCA project [Turner &Stevenson, 1991][Turner, 1994], whose goal is the creation of a robust, intelligent controller for AUVs for use in ocean science and CDPS systems.