There has been a great deal of work in mission-level control of autonomous and nearly-autonomous vehicles. Some of the major approaches include traditional AI planning mechanisms, rule-based reasoners, and reactive planning systems, including subsumption.
Traditional planners, even highly sophisticated ones such as [\protect\citenameWilkins, 1984] or [\protect\citenameVere, 1983], suffer from the drawbacks of overcommitment and lack of reactivity. They plan to complete detail before passing their plans off to an execution module (if one exists). They are isolated from changes in the problem-solving situation; when there is a change that the execution module cannot handle, the planner is re-invoked. This is unfortunate, by overcommiting to details, they will almost certainly have to repeat a substantial amount of reasoning as the world changes. In addition, the reasoner should be involved in most event handling, since it has the requisite knowledge needed to determine how the event impacts the mission and mission plan.
Rule-based reasoners can be used for planning (they can in general be used for whatever a general purpose computer can be, after all), though they are not the most natural choices. They divide actions up into independent rules; plans of action are, by their nature, interconnected sets of actions. By using rules, a reasoner may have behavior that can best be described as somewhat incoherent-that is, following no apparent plan of action. Worse, it is difficult to see how a rule-based reasoner can reuse its plans or can be given ``canned'' plans by humans. (This presents little problem for plan-based approaches, such as Orca.)
Reactive planning systems range from systems that are little more than versions of rule-based systems (e.g., [\protect\citenameAgre &Chapman, 1987]) or finite state machines (e.g., Brooks' subsumption architecture [\protect\citenameBrooks, 1986]) to planners that begin to look somewhat like Orca, such as [\protect\citenameGeorgeff &Lansky, 1987]. The former can suffer from the same problems as rule-based systems. In addition, they explicitly do not consider future states; thus, it is unclear how they would perform time-dependent actions or coordinate their activities with others. However, MIT Sea Grant's Odyssey AUV, which is subsumption-based, has achieved significant results [\protect\citenameBellingham et al., 1993].