Planning via heuristic search in a Petri net framework

Abstract

An artificial intelligence planning system's main components consist of a planner and a problem domain. The problem domain is the environment about which the planner reasons and on which it takes action. In the paper, a special type of extended input/output Petri net is defined and then used as the problem representation for a wide class of problem domains. A planning strategy is developed using results from the theory of heuristic search. In particular, using the developed Petri net framework and metric spaces, a class of heuristic functions that are both admissible and consistent for the A* algorithm is specified. The planning system architecture is discussed, and, as an illustration of the results, two simple planning problems are modeled and solved.<>