Stochastic Lyapunov Analysis for Consensus Algorithms with Noisy Measurements

Abstract

This paper studies the coordination and consensus of networked agents in an uncertain environment. We consider a group of agents on an undirected graph with fixed topology, but differing from most existing work, each agent has only noisy measurements of its neighbors' states. Traditional consensus algorithms in general cannot deal with such a scenario. For consensus seeking, we introduce stochastic approximation type algorithms with a decreasing step size. We present a stochastic Lyaponuv analysis based upon the total mean potential associated with the agents. Subsequently, the so-called direction of invariance is introduced, which combined with the decay property of the stochastic Lyapunov function leads to mean square convergence of the consensus algorithm.