Toward a Theory of Qualitative Reasoning about Plans

Abstract

Automated planning algorithms embody a theory of causality grounded in linking enabling effects of actions or initial world conditions to preconditions of subsequent actions. This model has several drawbacks when applied in the context of mixed-initiative planning. First, it requires comprehensive causal models that describe for every action its full set of preconditions and postconditions; in many application domains, such models will be unavailable. Second, it does not cover several forms of intraplan relations that a human may wish to document and reason with. To address these shortcomings, we define a qualitative framework for reasoning about plan structure. Specifically, we define a set of plan relations that characterize key interactions among plan components and a calculus for reasoning qualitatively about the effects of plan changes. We argue that our qualitative approach is much better suited to mixed-initiative planning than the causal link method.