Model-based Planning in Physical domains using SetGraphs

Abstract

This paper proposes a non-propositional representation framework for planning in physical domains. Physical planning problems involve identifying a correct sequence (plan) of object manipulations, transformations and spatial rearrangements to achieve an assigned goal . The problem of the ramification of action effects causes most current (propositional) planning languages to have inefficient encodings of physical domains. A simpler and more efficient representation is proposed, in which actions, goals and world state are modelled using ‘setGraphs’. A set Graph is an abstract data-structure able to capture implicitly the structural and topological constraints of a physical domain. Despite being model-based, the representation also allows the use of types and propositional furmulae to specify additional domain constraints. Experimental results obtained with a specific implementation of the representation indicate significant improvements in performance in all of the domains considered.