Classifying and Detecting Plan-Based Misconceptions for Robust Plan Recognition

My Ph.D. dissertation (Calistri 1990) extends traditional methods of plan recognition to handle situations in which agents have flawed plans.1 This extension involves solving two problems: determining what sorts of mistakes people make when they reason about plans and figuring out how to recognize these mistakes when they occur. I have developed a complete classification of plan-based misconceptions, which categorizes all ways that a plan can fail, and I have developed a probabilistic interpretation of these misconceptions that can be used in principle to guide a bestfirst–search algorithm. I have also developed a program called Pathfinder that embodies a practical implementation of this theory. Pathfinder is a probability-based plan-recognition system based on the A* algorithm that uses information available from a user model to guide a bestfirst search through a plan hierarchy. Most plan-recognition systems assume that the agent will have perfect plans, and sometimes this approach is acceptable. However, intelligent interfaces are specifically designed to interact with people, who can make mistakes. In fact, it is the least experienced people, the people who make the most mistakes, who need intelligent interfaces the most. The more mistakes a person is likely to make, the more important it is to be able to recognize and correct these mistakes. Ambiguity already shows up quite often in the traditional plan-recognition problem, but flawed plans are inherently ambiguous, and the number of alternative explanations is much larger than with correct plans. Although probabilistic methods have already been applied to plan recognition to handle ambiguity (Goldman and Charniak 1989), they have not addressed the A Robust PlanRecognition Algorithm