Incremental Plan Recognition in an Agent Programming Framework

Speaker: Yves Lesperance (York University)

In this talk, I will present a formal model of plan recognition for inclusion in a cognitive agent programming framework. The model is based on the Situation Calculus and the ConGolog agent programming language. This provides a very rich plan specification language. Our account also supports incremental recognition, where the set of matching plans is progressively filtered as more actions are observed. This is specified using a transition system account. The model also supports hierarchically structured plans and recognizes subplan relationships. I will talk about our prototype implementation and how it was used to recognize plans in a very simple "home activities" domain. I will also discuss work in progress on how we might represent concurrency in hypotheses about what plans have been executed and how we might deal with incomplete knowledge in recognition.

This is joint work with Alexandra Goultiaeva, who is now a Master's student at the University of Toronto.