Functional Strips: a More General Language for Planning and Problem Solving (preliminary Version)

The Strips language introduced by Fikes and Nilsson in 1971 has shaped most of the work in Planning. Its appeal can be explained by two reasons: Strips provides a compact language for representing actions that avoids the frame problem, and at the same time supports divide-and-conquer strategies that have been regarded as good for planning. In recent years, a number of extensions of Strips have been developed, yet they all share with Strips the following restriction: uent symbols, i.e. the symbols whose interpretation may change as a result of the actions, must be either relational or constant symbols. In this work we show that this restriction can be relaxed allowing for signiicantly more eecient representations and faster planning. More precisely, we move from the Strips language, that is based on relational uents, to Functional Strips, a language similar to Strips that is based on functional uents. Functional Strips can represent alìStrips problems', as well as problems that are dif-cult to encode or solve in Strips such problems involving resources. The advantage of a functional language over a purely relational one is that functions can be nested and hence can refer to objects without providing their explicit names. As a result, the number of arguments in action descriptions and the number of possible ground actions can be reduced substantially, something that is crucial in modern planners. For example, the 8-puzzle can be modeled by means of 4 ground actions, the Towers of Hanoi, with 3 ground actions, etc. Indeed, in many problems the number of ground actions gets reduced to the branching factor of the problem, and the state representations become as eecient as specialized representations. In the full version of the paper, we examine the performance of the 1 Heuristic Search Planner (HSP) used in the recent Planning Competition when extended to handle problems expressed in the new language.