Desires and Defaults: A Framework for Planning with Inferred Goals

In this paper, I will describe a formalism designed to integrate reasoning about desires with planning. One motive for such a formalism is the need to create a framework for reasoning about actions and change that provides for flexible reasoning about goals. The ideas rely on a crucial distinction between prima facie and all-things-considered attitudes. I model prima facie beliefs and desires as defaults, using the approach to nonmonotonic logic known as Default Logic. All-things-considered beliefs (representing actual epistemic commitments) and all-things-considered desires (representing goals) are selected by choosing an extension that is allowed by the logic. I show in the paper how to integrate these ideas with a planning formalism. The resulting logic, BDP, is capable of modeling a wide range of common-sense practical arguments, and can serve as a more general and flexible model for agent architectures.