2. State-Based Planning

Abstract

Planning is a major sub-discipline of AI. A plan is defined as a sequence of actions for transforming a given state into a state which fulfills a predefined set of goals. Planning research deals with the formalization, implementation, and evaluation of algorithms for constructing plans. In the following, we first (sect. 2.1.1) introduce a language for representing problems called standard Strips. Along with defining the language, the basic concepts and notations of planning are introduced. Afterwards (sect. 2.2) some extensions to Strips are introduced and situation calculus is discussed as an alternative language formalism. In section 2.3 basic algorithms for plan construction based on forward and backward search are introduced. We discuss complexity results for planning and give results for termination, soundness, and completeness of planning algorithms. In section 2.4 we give a short survey of well-known planning systems and an overview of concepts often used in planning literature and pointers to literature. Finally (sect. 2.5), pre-planning analysis and learning are introduced as approaches for acquisition of domain-specific knowledge which can be used to make plan contruction more efficient. The focus is on state-based algorithms, including universal planning. Throughout the chapter we give illustrations using blocks-world examples.