Organizing Plan Libraries in Subsumption Hierarchies: Specificity Based Plan Selection

Abstract : As the number of plans in a plan library grows, the importance of selecting a plan efficiently also increases. Previous planners have not addressed this issue in great detail, because they typically had only tens of plans - as domain models grow more and more specialized, and planners incorporate new macro-operators through learning, plan libraries can become more than a magnitude greater in size. This paper addressed the issue of organizing plan libraries can become more than a magnitude greater in size. This paper addressed the issue of organizing plan libraries in specificity based hierarchies. In particular, we look at how a plan library can be represented in a frame-based KL-ONE style system. Such systems offer powerful mechanisms to handle dynamically changing knowledge bases. In particular, the subsumption classifier mechanism in such systems, offers a simple and efficient means of indexing new plans and goals into the hierarchy. We illustrate the feasibility of this approach by using examples from an implemented system, and describe some of the other advantages that accrue from the use of such a framework for organizing large plan libraries. Plan libraries, Classification, Plan matching.