Summary: Learning Plan Rewriting Rules
Jos'e Luis Ambite, Craig A. Knoblock & Steven Minton
Information Sciences Institute
University of Southern California
4676 Admiralty Way, Marina del Rey, CA 90292, USA
fambite, knoblock, mintong@isi.edu
Abstract
Planning by Rewriting (PbR) is a new paradigm for ef­
ficient high­quality planning that exploits plan rewrit­
ing rules and efficient local search techniques to trans­
form an easy­to­generate, but possibly suboptimal,
initial plan into a high­quality plan. Despite the ad­
vantages of PbR in terms of scalability, plan qual­
ity, and anytime behavior, PbR requires the user to
define a set of domain­specific plan rewriting rules
which can be difficult and time­consuming. This pa­
per presents an approach to automatically learning the
plan rewriting rules based on comparing initial and op­
timal plans. We report results for several planning do­
mains showing that the learned rules are competitive

Source: Ambite, José Luis - Information Sciences Institute & Department of Computer Science, University of Southern California

Collections: Computer Technologies and Information Sciences