Prioritizing Planning Decisions in Real-World Plan Authoring

One promising direction for moving AI planning into the real world is to build systems that are more usercentric, in that they allow the user—not the system—to make many of the decisions necessary to create the final plan. However, since there are a large number of decisions to make in the course of producing a plan, shifting the responsibility for those decisions to the human planner runs the risk of overwhelming the human planner with too many choices. One approach to helping the human planner manage the large number of decisions is to automatically prioritize those decisions according to their importance or urgency in the current planning context. This paper describes two methods for automatically prioritizing planning decisions. One is a commitment-based approach, which prioritizes decisions according to the number of future decisions they eliminate from the planning process. The other is an experience-based approach, which prioritizes decisions according to the order in which they have been performed in previous planning sessions. Both approaches have been implemented in PASSAT, a plan-authoring system in which users construct and modify plans interactively using a library of templates.