Ron Petrick's Webpages

Workshop paper

Improving Planner Performance in Grid Worlds with Macro Actions, M. Crosby and R. Petrick, Proceedings of the International Cognitive Robotics Workshop (CogRob 2014) at ECAI 2014, 2014.

[ bib | pdf ]

Abstract

In this paper we explore a class of grid world planning domains that models high-level multi-robot navigation in confined spaces, and that gives rise to certain problem instances over which some modern planning techniques perform surprisingly poorly. We show that the inclusion of macro actions, inspired by techniques that humans use to solve similar problems, allows planners to find solutions in some cases where they would otherwise fail. We then show that the pattern exploited to create the macro actions could potentially be exploited in the majority of planning domains used in the International Planning Competition.