Showing papers by "Paul Morris published in 1984"

A resource oriented formalism for plan generation (artificial intelligence, problem solving)

Abstract: A formulation of planning based on multisets is introduced. Actions are represented by a simple well-behaved formalism which is related both to Petri nets and to predicate logic. It is inherently resource and quantity oriented and more accurately models constraints imposed by resource limitations. The consequences of this for plan generation are explored. The formalism is simple enough to allow theoretical issues to be investigated. Parallel plans are represented by a concise graphical structure with the property that a syntactically complete plan is necessarily correct. Abstract algorithms for plan generation with provable completeness properties are exhibited. Techniques for recognizing and escaping from vicious circles are studied. These turn out to also have the effect of excluding certain correct but nonsensical plans, and allow the automatic derivation of planning-time checks on the appropriateness of specific operators, but cause incompleteness in some domains. Novel methods for efficient planning that arise naturally from the formalism are described. Cut and splice operations on plan graphs permit the order of goal selection to be based on efficiency rather than correctness considerations. The multiset based formalism is particularly suited to developing the notion of an invariant, or conservation law. Invariants are exploited in a technique called goal targeting which provides an early outline of the final plan. This helps to eliminate misdirected search and facilitates certain intelligent goal selection and variable binding mechanisms.