Expressive Equivalence of Planning Formalisms

A concept of expressive equivalence for planning formalisms based on polynomial transformations is de ned. It is argued that this de nition is reasonable and useful both from a theoretical and from a practical perspective; if two languages are equivalent, then theoretical results carry over and, more practically, we can model an application problem in one language and then easily use a planner for the other language. In order to cope with the problem of exponentially sized solutions for planning problems an even stronger concept of expressive equivalence is introduced, using the novel ESP-reduction. Four di erent formalisms for propositional planning are then analyzed, namely two variants of STRIPS, ground TWEAK and the SAS formalism. Although these may seem to exhibit di erent degrees of expressive power, it is proven that they are, in fact, expressively equivalent under ESP reduction. This means that neither negative goals, partial initial states nor multi-valued state variables increase the expressiveness of `standard' propositional STRIPS. This article is published in Arti cial Intelligence, vol. 76(1{2):17{34, 1995, ARTINT 1234.