Planning in Description Logics: Deduction versus Satisfiability Testing

Description Logics (DLs) are formalisms for tax-onomic reasoning about structured knowledge. Adding the transitive closure of roles to DLs also enables them to represent and reason about actions and plans. The present paper explores several essentially diierent encodings of planning in Description Logics. We argue that DLs represent an ideal framework for analysing and comparing these approaches. Thus, we have identiied two essentially diierent deductive encodings (a \causal" and a \symmetric" one), as well as a satissability-based approach. While the causal encoding is more appropriate for reasoning about precondition-triggered causal events, the symmetric encoding is more amenable to reasoning about possible outcomes of courses of actions without actually executing them (while allowing both progression and regression). In the deductive approaches, the existence of a plan corresponds to an inconsistency proof rather than to a model of some formula. Viewing planning as satissability testing addresses this problem by reducing planning to model construction.