Improving MaxSAT-Reordered Plans via Block Deordering

Abstract

Partial-order plans (POPs) provide greater flexibility during execution compared to sequential plans due to their least commitment nature. Optimizing flexibility in a POP involves strategies such as plan deordering, which eliminates unnecessary action orderings, and plan reordering which modifies action orderings arbitrarily. Though traditional plan deordering techniques, such as EOG (explanation-based order generalization), can efficiently find partial orderings in polynomial time, they lack optimality guarantees. This limitation prompts MaxSAT reorderings to encode the optimization of a POP’s orderings as a partial weighted MaxSAT problem. To further elevate the flexibility of the MaxSAT solutions, this work introduces an algorithm that employs block deordering, a distinct form of plan deordering that consolidates coherent actions into blocks, on top of MaxSAT reorderings. Our experiments with benchmark problems from the International Planning Competitions demonstrate that our algorithm not only makes significant enhancements to the satisfiable MaxSAT reordered plans but also takes it a step further by improving the optimal reordered plans in mere seconds.