Towards probabilistic plan management

Abstract

In temporally uncertain domains, taking uncertainty into account while planning leads to problems with scalability. One alternative to this is to plan deterministically and replan when execution deviates from schedule. In large, complex problems, however, replanning during execution can be prohibitively expensive. To address this, we have developed a general plan management framework called Probabilistic Plan Management (PPM). PPM probabilistically limits how far in the future it is necessary to consider tasks while repairing and replanning during execution. PPM also decides whether to replan based on the probability that a violated constraint will occur in execution, not on the presence of a conflict in the plan. These features decrease replanning during execution while ensuring that the quality of execution does not unduely suffer. In this paper, we describe our approach and discuss results in simulation that show large savings in the total time spent replanning during execution.