Most work on AI planning has focused on the development of fully automated methods for generating plans that satisfy user-speci ed goals. However, users in many domains want the ability to in uence the nature of the solutions that are generated. With the objective of fostering increased user participation in the planning process, this paper presents an HTNbased framework for the abductive completion of plan sketches. Within this framework, user-supplied outlines of partial plans possibly spanning multiple abstraction levels are interpreted and completed. The processing of plan sketches employs an initial abductive plan recognition phase to formulate candidate sets of intended user goals, followed by a plan re nement stage that generates sketch-compliant nal plans for those goals. A prototype sketch-based planner based on this approach has been implemented and applied to a crisis action planning domain.
Karen L. Myers