Efficient Communication Infrastructures for Distributed Control and Decision Making in Networked Stochastic Systems

In networked systems, groups of agents achieve certain objectives via interaction at local levels in a decentralized manner. The performance of such systems is determined by the communication infrastructure of the network as well as the system dynamics. The interdependence of agents in a networked system is often modeled by graphs. We study the interdependence of communication and collaboration graphs in a networked system in the context of a coordination control and decision making problems. We model the decision on whether to cooperate or not in a group effort as a result of a series of two-person games between agents and their neighbors. The payoff of each agent is computed as the sum of the agent’s payoffs from each of these games. Since coordination games have more than one equilibrium point, the problem is then which equilibrium point will the agents choose and whether they will settle on a Pareto-optimal equilibrium point. We consider a behavior learning algorithm and study its effect on the emergence of a collaboration graph. We also study the effect of the communication network topology on the convergence speed of the scheme.