Approximation Algorithms and Heuristics for Classical Planning

Abstract

Automated planning has been an active area of research in theoretical computer science and Artificial Intelligence (AI) for over 40 years. Planning is the study of general purpose algorithms that accept as input an initial state, a set of desired goal states, and a planning domain model that describes how actions can transform the state. The problem is to find a sequence of actions that transforms the initial state into one of the goal states. Planning is widely applicable, and has been used in such diverse application domains as spacecraft control [MNPW98], planetary rover operations [BJMR05], automated nursing aides [MP02], image processing [GPNV03], computer security [BGHH05] and automated manufacturing [RDF05]. Planning is also the subject of continued and lively ongoing research. In this chapter, we will present an overview of how approximations and related techniques are used in automated planning. We focus on classical planning problems, where states are conjunctions of propositions, all state information is known to the planner, and all action outcomes are deterministic. Classical planning is nonetheless a large problem class that generalizes many combinatorial problems including bin-packing ∗Universities Space Research Association †The authors gratefully acknowledge Sailesh Ramakrishnan, Ronen Brafman and Michael Freed for reviewing our early drafts