Temporally-expressive planning as constraint satisfaction problems
Yuxiao Hu · 2007
Due to its important practical applications, temporal planning is of great research interest in artificial intelli-gence. Yet most of the work in this area so far is limited in at least two ways: it only considers temporally sim-ple domains and it has restricted decision epochs as the potential happening time of actions. Because of these simplifying assumptions, existing temporal planners are in fact not complete. In this paper, we focus on these limitations, and pro-pose an alternative view of temporal planning by inves-tigating a new declarative semantics of PDDL. We then show a natural encoding of this semantics in a constraint programming setting. It turns out that this encoding uni-fies planning and scheduling, and captures most of the temporal expressiveness of PDDL. The resulting CSP-based temporal planner can solve more general planning problems than the current state-of-the-art.