Means-End Plan Recognition - Towards a Theory of Reactive Recognition

This paper draws its inspiration from current work in reactive planning to guide plan recognition using \plans as recipes". The plan recognition process guided by such a library of plans is called means-end plan recognition. An extension of dynamic logic, called dynamic agent logic, is introduced to provide a formal semantics for means-end plan recognition and its counterpart, means-end plan execution. The operational semantics, given by algorithms for means-end plan recognition, are then related to the provability of formulas in the dynamic agent logic. This establishes the relative soundness and completeness of the algorithms with respect to a given library of plans. Some of the restrictive assumptions underlying means-end plan recognition are then relaxed to provide a theory of reactive recognition that allows for changes in the external world during the recognition process. Reactive recognition, when embedded with the mental attitudes of belief, desire, and intention, leads to a powerful theory of integrated reactive planning and recognition. The primary contribution of this paper is in laying the foundations for such an integrated theory.