The Design of Autonomous Robot Representation

This paper presents a design methodology for representation systems used by the “perception/action oriented ” portions of agent architectures in “dynamic domains”. By “perception/ action oriented ” we mean the portion of the architecture that is concerned with operating in the physical world, not the portion concerned with generating courses of action, by inferencing or planning. By “dynamic domains ” we mean task domains in which current sensor values are thought to be a more accurate representation of the world than previous sensor data. We present a design methodology for developing representation systems for autonomous agents in such domains and describe three physical agents that have been implemented using this methodology. We also present criteria for judging effective systems of representation and discuss how our agent’s representation