ECAI 2004 Conference Paper

[full paper]

NGP: Numerical Graph Planning

Joseph Zalaket, Guy Camilleri

Most of the real world problems involve numerical handling. New planning systems such as FF-Metric and SAPA are able to handle numerical resources, but in most cases the resources are only used as auxiliary numerical constraints added to the symbolic planning domain. The extension to MIPS to solve numerical domains uses as for FF-Metric and SAPA two different types of heuristic one for the propositional part of the planning problem and another one for its numerical information. In this paper we present a new planning system NGP (Numerical Graph Planning) that is able to solve totally numerical and/or symbolic planning domains. We propose a new action representation to support numerical conditions and effects, where we allow a non-restricted function application for numerical update. NGP guides its search using a heuristic derived from a planning graph construction. In NGP we instantiate symbolic actions parameters from the outset for only one time during the planning process, but we progressively instantiate the numerical actions parameters every time a new graph layer is reached. NGP explicitly represents the possible range of numerical parameters for the numerical functions of the domain.

Keywords: planning, heuristic search

Citation: Joseph Zalaket, Guy Camilleri: NGP: Numerical Graph Planning. In R.López de Mántaras and L.Saitta (eds.): ECAI2004, Proceedings of the 16th European Conference on Artificial Intelligence, IOS Press, Amsterdam, 2004, pp.1115-1116.