Plan representation for robotic agents

Abstract

Most robotic agents cannot fully exploit plans as resources for better problem-solving performance because of imminent limitations of their plan representations. In this paper we pro-pose plan representations that are, for a given job, represen-tationally and inferentially adequate and inferentially and ac-quisitionally efficient. We state what these properties mean in the context of robotic agents and describe how plan represen-tations can be designed to satisfy them. The proposed plan representations have been successfully employed in several longterm experiments on autonomous robots.