Evaluating the Cost of Employing LPs and STPs in Planning: Lessons Learned From Large Real-Life Domains

Elad Denenberg, Amanda Coles, Derek Long

Keywords: Planning, Planning Applications, Temporal Planning, Numeric Planning, Hybrid Planning

Abstract: When solving real-life problems we often encounter issues that are not captured by academic benchmark domains. In this paper we consider an application problem, representative of a class of real-world problems that have interesting properties: long solution plans with many temporal/numeric constraints. We identify a number of limitations of a popular family of planners in solving these problems. This family of planners perform Forward Search and call an LP solver multiple times at every state to check for consistency, and to set bounds on the numeric variables in order to determine action applicability. These checks during search allow the pruning of branches; however, they do carry computational cost. In this paper we investigate and analyse this trade-off, with particular reference to our class of application problems, and show that adapting the planners to call the LP solver less often, and using a cheaper consistency check at each state, can improve performance.

Author Identity Visibility: Reveal author identities to reviewers