Domain-independent, or knowledge-sparse, planning has limited practical appli-cation because of the failure of brute-force search to scale to address real prob-lems. However, requiring a domain engineer to take responsibility for directing the search behavior of a planner entails a heavy burden of representation and leads to systems that have no general application. An interesting compromise is to use domain analysis techniques to extract features from a domain description that can exploited to good effect by a planner. In this chapter we discuss the process by which generic patterns of behavior can be recognized in a domain, by automatic techniques, and appropriate specialized technologies recruited to assist a planner in efficient problem solving in that domain. We describe the in-tegrated architecture of STAN5 and present results to demonstrate its potential on a variety of planning domains, including two that are currently beyond the problem-solving power of existing knowledge-sparse approaches.
|Title of host publication||Exploring Artificial Intelligence in the New Millennium|
|Number of pages||35|
|Publication status||Published - 2002|
|Name||Morgan Kaufmann Series in Artificial Intelligence|
Long, D., Fox, M., Lakemeyer, G. (Ed.), & Nebel, B. (Ed.) (2002). Planning with generic types. In Exploring Artificial Intelligence in the New Millennium (pp. 103-138). (Morgan Kaufmann Series in Artificial Intelligence).