Distributed Kalman filtering for multiagent systems

For naturally distributed systems, such as multiagent systems, the construction and tuning of a centralized observer may be computationally expensive or even intractable. An important class of distributed systems can be represented as cascaded subsystems. For this class of systems, observers may be designed separately for the subsystems. If the subsystems are linear, the Kalman filter provides an efficient means to estimate the states, so that it minimizes the mean squared estimation error. Kalman-like filters may be used for the whole system or the individual subsystems. In this paper, both a theoretical comparison and simulation examples are presented. The theoretical results show that the distributed observers, except for special cases, do not minimize the overall error covariance, and so the distributed observer system is suboptimal. However, in practice, the performance achieved by the cascaded observers is comparable and in certain cases outperforms that of the centralized one. Moreover, a distributed observer system leads to increased modularity, reduced complexity, and lower computational costs.