The author proposes a knowledge-level model for keyhole plan recognition. This model supports knowledge acquisition needed for plan recognition by making explicit and structuring, at an implementation-independent level, the relationships among the types of knowledge and the elements of the reasoning process involved in that knowledge.

Plan recognition involves interpreting an agent's intentions by ascribing goals and plans to it. In treating plan recognition as a problem-solving activity, this article proposes a knowledge-level model of this task for a case of noncooperative recognition that uses low-level data in a dynamic environment. Such a model contributes to our understanding of plan recognition by making explicit, at an implementation-independent level, the problems a plan recognizer must solve, the reasoning for solving them, the knowledge needed to perform this reasoning, and the constraints imposed by the plan recognition environment. For modeling and prototyping, my colleagues at the University of Paris-Sud and I first applied this model to the domain of driving a car. It has also guided knowledge acquisition and design activities in the construction of a plan recognition module for integration in a military command, control, communication, and intelligence (C3I) system.