A Temporal Relaxed Planning Graph Heuristic for Planning with Envelopes

Abstract

When planning in temporal domains with required concurrency, envelopes arise where one or more actions need to occur within the execution of another. Starting an envelope action gives rise to an implicit relative deadline: all of the actions that need to occur within the envelope must complete before it ends. Finding effective heuristic guidance in these domains is challenging: the heuristic must not only consider how to reach the goals, but identify when it is not possible to achieve these implicit deadlines to avoid fruitless search. In this paper, we present an adaptation of a Temporal Relaxed Planning Graph heuristic, that accounts for dependencies between facts and actions in the relaxed planning graph; and the envelopes that are open in the state being evaluated. Results show that our new heuristic significantly improves the performance of a temporal planner on benchmark domains with required concurrency.