On procedure recognition in the Situation Calculus

Abstract

The aim of our ongoing research is to give a method to construct intelligent tutoring systems for agents who are executing typical procedures in dynamic environments based on a logical framework. Typical procedures are similar to plans in the sense that they describe the actions an agent should execute to achieve a certain goal. In this paper we address what we consider is the first step toward the construction of this kind of system: procedure recognition. We formalize what it means that an agent is performing a procedure in the Situation Calculus, a logical first-order language extended with induction. Based upon this formalization, we give two different implementations. The first, which is directly based in our formalization, is proved to be quite inefficient. The second, significantly more efficient, arises from a logical reformulation of the original formalization. Procedures are represented through CONGOLOG programs, a logical interpreted language based on the Situation Calculus.