Leveraging Imperfect Explanations for Plan Recognition Problems.

Open environments require dynamic execution of plans where agents must engage in settings that include, for example, re-planning, plan reusing, plan repair, etc. Hence, real-life Plan Recognition (PR) approaches are required to deal with different classes of observations (e.g., exogenous actions, switching between activities, and missing observations). Many approaches to PR consider these classes of observations, but none have dealt with them as deliberated events. Actually, using existing PR methods to explain such classes of observations may generate only so-called imperfect explanations (plans that partially explain a sequence of observations). Our overall approach is to leverage (in the sense of plan editing) imperfect explanations by exploiting new classes of observations. We use the notation of capabilities in the well-known Belief-Desire-Intention (BDI) agents programming as an ideal platform to discuss our work. To validate our approach, we show the implementation of our approach using practical examples from the Monroe Plan Corpus.