A long-range AUV has a degree of autonomy greater than its short-range relatives, since it will often be out of touch with humans for extended periods of time. The longer the mission, the more important the effects of uncertainty and incomplete knowledge about the environment are. Consequently, where some short-range missions can be accomplished by downloading detailed plans to the AUV before launch, an LRAUV treated the same way would likely fail: the circumstances unforeseen by the users will likely arise, making the downloaded plans obsolete. Instead, an LRAUV needs onboard intelligence sufficient to create, monitor, and change plans as its world changes. That is, it needs an onboard mission planner.
There is currently much confusion over the term ``planning'' in the AI literature. It has been given meanings ranging from the non-reactive creation of detailed actions for later execution [e.g.,]strips to the highly reactive use of simple stimulus-response ``rules'' to respond to current conditions in the environment [e.g.,]agre87. We take a middle-ground approach, using the term in a broad sense to mean the creation of plans of actions at some level of detail, while interleaving execution of the plan with planning or of reacting to stimuli during planning or execution. This is essentially the use of the term in much of the AI literature of reactive planning [e.g.,]georgeff87.
The mission planner for an LRAUV needs to be able to create plans from scratch (i.e., from primitive operators comprising the AUV's operator set and its own, private, operators), but it also should be able to reuse plans it has been given or has used before for efficiency. It should not, however, create completely detailed plans for long periods into the future; in fact, for many situations, it will lack the necessary sensor information, even if it were desirable to do so. To do so invites constant replanning as the situation changes unpredictably. Instead, it should have an overall idea of the course of the mission, complete with spatiotemporal ``anchors'' to important mission events, but should plan the details only as needed. One way to do this is to execute directly from a hierarchical plan, expanding the only those portions of the plan needed to choose what to do next.