A distributed, decision-theoretic control system for a mobile robot

Abstract

A model is described that divides an agent architecture into five components (percepts, decomposition, strategies, arbitration, actions). Using this model and a decision-theoretic perspective on the agent's behavior, the resultant agent is both flexible and rational. A mobile robot control system that uses the agent architecture model is described in some depth. The mobile robot operates in a simulated, 2-D environment with fixed and moving obstacles and a limited and noisy vision sensor. A formal, comparative analysis of agent architecture classes concludes this paper.