R. Wade Brittain and Bruce D'Ambrosio
Due to limited power supplies, finite dust storage, poor location sensing, and random failures caused by sucking up numerous small objects, autonomous vacuum cleaners require accurate and timely information about environmental parameters and the state of internal components. The best methods for providing such estimates are Bayesian. The recent advent of compact, efficient probabilistic representations has made the implementation of these methods not only more tractable, but has also extended the range and scope of architectures in which they can be implemented. We explore the anomalies of the vacuuming domain and illustrate how Bayesian techniques can be applied to reduce their effects on agent performance.