Anagent can performerroneous actions. Despite such errors, one might want to understand what the agent tried to achieve. Suchunderstanding is important, for example, in intelligent tutoring and expert critiquing systems. In intelligent tutoring, feedbackto the agent--i.e, the student--should focus on which actions madesense and which did not. To achieve such understanding, wepropose a combinedplanning and plan recognition approach. This paper focuses on plan-based plan recognition. Wedescribe a generic recognition algorithm and showhowit can be used for plan recognition and critiquing. Theapproachis illustrated using a real-world application in the area of multi-mediatutoring of ship crisis management.
Ole J. Mengshoel, David C. Wilkins