On the Plan-Library Maintenance Problem in a Case-Based Planner Alfonso Emilio Gerevini 1 , Anna Roub´ ıˇ ckov´ a 2 , Alessandro Saetti 1 , and Ivan Serina 1 1 Dept. of Information Engineering, University of Brescia, Brescia, Italy 2 Faculty of Computer Science, Free University of Bozen-Bolzano, Bolzano, Italy {gerevini,saetti,serina}@ing.unibs.it, anna.roubickova@stud-inf.unibz.it Abstract. Case-based planning is an approach to planning where pre- vious planning experience stored in a case base provides guidance to solving new problems. Such a guidance can be extremely useful when the new problem is very hard to solve, or the stored previous experience is highly valuable (because, e.g., it was provided and/or validated by human experts) and the system should try to reuse it as much as pos- sible. However, as known in general case-based reasoning, the case base needs to be maintained at a manageable size, in order to avoid that the computational cost of querying it excessively grows, making the entire approach ineffective. We formally define the problem of case base mainte- nance for planning, discuss which criteria should drive a successful policy to maintain the case base, introduce some policies optimizing different criteria, and experimentally analyze their behavior by evaluating their effectiveness and performance. 1 Introduction It is well known that AI planning is a computationally very hard problem [9]. In order to address it, over the last two decades several syntactical and structural restrictions that guarantee better computational properties have been identified (e.g., [3,4]), and various algorithms and heuristics have been developed (e.g., [7,16]). Another complementary approach, that usually gives better computa- tional performance, attempts to build planning systems that can exploit ad- ditional knowledge not provided in the classical planning domain model. This knowledge is encoded as, e.g., domain-dependent heuristics, hierarchical task networks and temporal logic formulae controlling the search, or it can be auto- matically derived from the experiences of the planning system in different forms. Case-based planning (e.g., [8,15,17,21]) follows this second approach and con- cerns techniques that improve the overall performance of the planning system by reusing its previous experiences (or “cases”), provided that the system frequently encounters problems similar to those already solved and that similar problems have similar solutions. If these assumptions are fulfilled, a well-designed case- based planner gradually creates a plan library that allows more problems to be solved (or higher quality solutions to be generated) compared to using a clas- sical domain-independent planner. Such a library is a central component of a S.J. Delany and S. Onta˜ n´on (Eds.): ICCBR 2013, LNAI 7969, pp. 119–133, 2013. c  Springer-Verlag Berlin Heidelberg 2013