Consistency driven planning

Abstract

This paper describes a novel approach to planning. The presented algorithm is based on a consistency maintaining procedure for computing possible worlds out of given worlds and applications of operators. Worlds are represented by facts, rules, and consistency constraints. In order to avoid the frame and qualification problems we state neither frame axioms nor qualification axioms. Instead, general consistency constraints are used. As a result an execution of an action which asserts its postconditions to the current state of the world may result in inconsistency. A repair mechanism then generates possible changes (repairs) to the inconsistent world such that the resulting world describes the actual consistent state of affairs, i. e. a possible world after the execution of an action.

The generated repairs serve two purposes: as they describe possible worlds in which the action's postconditions hold, they can be used to reason about these worlds and eliminate those which have undesired properties. Moreover, the repairs are used to guide the linear planner such that the generated plans lead to the selected possible world.

Part of the research was performed while the author was visiting the Dept. of AI, University of Edinburgh, Scotland. This visit was supported by the Commission of European Communities under grant ERBSCI*CT000301 — (SC1000301).