Generating Temporally Contingent Plans

Uncertainty applies to many aspects of planning problems. Much research has been done to deal with problems where actions have uncertain effects. In reality, many planning problems also involve actions with uncertain durations, but this type of uncertainty has not been widely studied in planning until the recent development of several planners which incorporate durational uncertainty. Additionally, theoretical work has been done on characterizing the level of controllability in plans involving actions with uncertain durations. We have developed an approach for finding temporally contingent plans, i.e., plans with branches that are based on the duration of an action at execution time. More specifically, the problems we are studying satisfy the following criteria: (1) there is more than one solution plan, (2) solution plans are ranked by a metric that is not fully based on makespan, (3) actions have uncertain durations, (4) the start and/or end times of some actions are constrained, and (5) as actions require more time to complete, plans that are judged highly by the metric become invalid. We describe our approach for determining the time points that cause an unsafe situation and for inserting temporal contingency branches. Experimental results with both sequential and parallel plans are discussed.