Specifying and updating plan libraries for plan recognition tasks

The authors present research in plan recognition which enables systems to be built which will perform plan recognition with plan libraries that are incompletely specified and later updated by a user. The algorithms build on Kautz's model of plan recognition, but introduce an assumption-based truth maintenance system to update the set of candidate plans, adjusting to the new library information. Plan recognition is achieved via local repair driven by the changes to the library, rather than by complete recalculation from the new library. This approach is considered desirable in practical applications where plan libraries may be very large. A graphical tool is presented which allows a user to specify a plan library, observations and updates, and displays candidate plans to the user as output. This tool is in the development stage. This work addresses the problems faced by users in application areas which employ plan recognition, allowing for specifications and modifications to be supplied in stages with plan recognition continuing at minimal repair cost.<>