Distributed estimation systems

This paper considers the distributed estimation problem by a set of agents connected by an arbitrary communication network. The agents communicate conditional probabilities of the random state over the network. From these conditional probabilities, each agent then tries to re-construct the conditional probability given all the measurements if these were communicated instead of the probabilities. It is discovered that in general the agents have to remember some of the past conditional probabilities and may even have to request additional information. A method for generating the fusion algorithm for each agent based on the network structure is presented and applied to some examples. The results are applicable to both dynamic and static states.