Asymptotic Agreement in a Class of Networked Kalman Filters With Intermittent Stochastic Communications

A sensor network with time varying communication links is considered. Data sharing among nodes is described by a stochastic time-varying network topology. Individual estimates are determined by local noisy measurements, and by the independent opinion pool data fusion scheme adapted to the distributed network structure. The sequences describing communication events between pairs of agents are modeled as Markov chains. It is shown that if communication events are recurrent in the stochastic sense, then the error covariance matrices associated with individual estimates are asymptotically independent of the sensor node. This result applies to coupled estimation and motion control with networked systems, for example to establish synchronization results in certain networks that are disconnected at any given time instant.