Janae N. Foss and Nilufer Onder
In classical planning assumptions are made that ignore important aspects of the real world. As a result, many planners have been written that extend classical planning in various ways. Often times these extensions to classical planning incorporate aspects of scheduling such as optimization and reasoning about resources and time. For example, there has been considerable research with planners that allow durative actions. Uncertainty is an aspect of the real world that is being studied in both planning and scheduling communities. (Dearden et al. 2003) lists a classification of planners that deal with uncertainty. Most of the research in uncertainty has dealt with uncertain action effects. However we are interested in actions with uncertain durations. The work in planners that assume action duration is uncertain has focused on finding conservative plans that are safe regardless of how long the actions in the plan take. The main disadvantage of the conservative approach is that opportunities are often missed. To help resolve this issue, we are writing a planner that takes a different approach by generating temporally contingent plans. These plans include branches that are based on the duration of actions at execution time. The advantage to this approach is that the conservative plan still exists, but only as a contingency branch, allowing a more vigorous plan to be executed when time allows.