A case-based approach to heuristic planning

Abstract

Most of the great success of heuristic search as an approach to AI Planning is due to the right design of domain-independent heuristics. Although many heuristic planners perform reasonably well, the computational cost of computing the heuristic function in every search node is very high, causing the planner to scale poorly when increasing the size of the planning tasks. For tackling this problem, planners can incorporate additional domain-dependent heuristics in order to improve their performance. Learningbased planners try to automatically acquire these domaindependent heuristics using previous solved problems. In this work, we present a case-based reasoning approach that learns abstracted state transitions that serve as domain control knowledge for improving the planning process. The recommendations from the retrieved cases are used as guidance for pruning or ordering nodes in different heuristic search algorithms applied to planning tasks. We show that the CBR guidance is appropriate for a considerable number of planning benchmarks.

Progress in Case-Based Planning

“…There are several ways to identify such cases. Associative retrieval uses the features of the cases to assess the suitability of a case, such as by using a similarity metric [De la Rosa et al 2013;Hanks and Weld 1995]. Hierarchical retrieval uses a hierarchical ordering of the features, proceeding from the more general to the more specific ones, such as indexing-based retrieval [Serina 2010;Tonidandel and Rillo 2002;Veloso 1994].…”

“…Thus, the first step of the retrieval phase often consists of fitting the objects of the two problem descriptions to each other as closely as possible. Depending on the implementation, the task may be realized either as unification [Hanks and Weld 1995;Kambhampati and Hendler 1992;Köhler 1996] or as object matching [De la Rosa et al 2013;Nebel and Köhler 1995;Serina 2010].…”

“…Others perform more dynamic plan changes using a deterministic [Likhachev and Koenig 2005;Van der Krogt and de Weerdt 2005] or stochastic [Borrajo and Veloso 2012;Fox et al 2006] sequence of plan modification operators that try to repair the failed plan. Alternatively, the system can try to replay decisions that were successful in the search for the solution found in the case base, yielding a derivational adaptation [De la Rosa et al 2013;Veloso 1994]. Another alternative consists of using a hybrid approach that combines several of the techniques mentioned previously.…”

“…CABALA [De la Rosa et al 2013] was designed to be a recommendation system for heuristic planners, like FF. It incorporates a full CBP cycle, from retrieval to retaining new cases.…”

“…The footprint is computed from the relaxed plan, focusing on each specific object. Details can be found in the work of De la Rosa et al [2013]. This serves to differentiate the best case for each hoist.…”