Action, Abduction And Plan Recognition

In the forthcoming distributed autonomous robotic systems it will be useful for a robot to recognize other robots' goals and plans from visual information. Till now, much emphasis has been given to plan inference. This paper is about goal recognition: having recognised a plan (may be after the entire plan has been performed), try to recognise which can be the actor's reasons for the plan to be performed. If the actor's planner possesses sufficient inferential capabilities, then goal recognition is not a trivial question. This paper shows that, under simple hypotheses on the nature of the planner that guides an actor's behaviour, an observer can recognize the actor's goal by means of a simple clause-based abductive reasoning. Furthermore, the paper shows how goal recognition can be regarded as a useful step in plan inference. This results refer to the prototypical state-based STRIPS planner