Evaluation Strategies for Planning as Satisfiability

Abstract. We investigate different evaluation strategies for planning problems represented as constraint satisfaction or satisfiability problems. The standard evaluation strategy, evaluating the formulae by sequentially increasing the length one step at a time, guarantees that a plan corresponding to the first satisfiable formula is found first, yet this is often not the best possible strategy in terms of runtime. We present evaluation strategies based on parallel or interleaved evaluation of several formulae and show that with many problems this leads to substantially improved runtimes, sometimes several orders of magnitude. The cost of the improved runtimes is a possible decline in plan quality because an optimality guarantee of the standard evaluation strategy is lost.

• Type: Conference
• Year: 2004
• Where: ECAI
• Authors: Jussi Rintanen
• Keywords: Artificial Intelligence, ECAI 2004, Evaluation Strategies, Satisfiability Problems, Standard Evaluation Strategy