Online identification of useful macro-actions for planning
Andrew J. Coles, Maria Fox, Amanda J. Smith · 2007
This paper explores issues encountered when performing on-line management of large collections of macro-actions gen-erated for use in planning. Existing approaches to manag-ing collections of macro-actions are designed for use with of-fline macro-action learning, pruning candidate macro-actions on the basis of their effect on the performance of the plan-ner on small training problems. In this paper we intro-duce macro-action pruning techniques based on properties of macro-actions that can be discovered online, whilst solving only the problems we are interested in. In doing so, we re-move the requirement for additional training problems and offline filtering. We also show how search-time pruning tech-niques allow the planner to scale well to managing large col-lections of macro-actions. Further, we discuss the properties of macro-actions that allow the online identification of those that are likely to be useful in search. Finally, we present results to demonstrate that a library of macro-actions man-aged using the techniques described can give rise to a sig-nificant performance improvement across a collection of do-mains with varied structure. 1