Policy Generation for Continuous-time Stochastic Domains with Concurrency

We adopt the framework of Younes, Musliner, & Simmons for planning with concurrency in continuous-time stochastic domains. Our contribution is a set of concrete techniques for policy generation, failure analysis, and repair. These techniques have been implemented in TEMPASTIC, a novel temporal probabilistic planner, and we demonstrate the performance of the planner on two variations of a transportation domain with concurrent actions and exogenous events. TEMPASTIC makes use of a deterministic temporal planner to generate initial policies. Policies are represented using decision trees, and we use incremental decision tree induction to efficiently incorporate changes suggested by the failure analysis.