Towards a unified view of AI planning and reactive synthesis

Abstract

Automated planning and reactive synthesis are wellestablished techniques for sequential decision making. In this paper we examine a collection of AI planning problems with temporally extended goals, specified in Linear Temporal Logic (LTL). We characterize these so-called LTL planning problems as two-player games and thereby establish their correspondence to reactive synthesis problems. This unifying view furthers our understanding of the relationship between plan and program synthesis, establishing complexity results for LTL planning tasks. Building on this correspondence, we identify restricted fragments of LTL for which plan synthesis can be realized more efficiently.