Knowledge and Regularity in Planning
J. A. Allen · 1992
this paper we present Talus, a system that can use domain-specific plan knowledge in each of these modes. This lets us compare the three approaches experimentally to determine the conditions under which each is most appropriate. We begin by describing Talus' representation of plan knowledge, followed by its behavior when using this knowledge, analogically, as macrooperators, and as search-control rules. After this we propose hypotheses about the relation among these methods, characteristics of planning domains, and the behavior of the methods in such domains. We then report experimental studies that test these hypotheses. In closing we draw some tentative conclusions and outline directions for future work.