Flexible goal recognition via graph construction and analysis

Instead of using a plan library, the recognizer introduced in this paper uses a compact structure called flexible to represent goals, actions and states of the world. This method doesn't suffer the problem of acquisition and hand-coding a larger plan library as traditional methods do. The recognizer also extends classical methods in two directions. First, using flexible goals and actions via fuzzy sets, the recognizer can recognize goals even when the agent has not enough domain knowledge. Second, the recognizer offers a method for assessment of various plan hypothesis and eventual selection good ones. Since the recognizer is domain independent the method can be adapted in almost every domain. Empirical and theoretical results also show the method is efficiency and scalability.