Comparison of centralized and decentralized linear-quadratic-Gaussian path control problems

The authors solve centralized and decentralized aerospace control problems explicitly accounting for sensor-to-sensor estimate cross-correlation due to the dynamics model and sensor-to-sensor measurement cross-correlation. A linear multinode network with a quadratic performance criterion was treated. The case of nonstationary measurement noise requires the parallel real-time calculation of intersensor estimate error and/or measurement cross-covariances. When steady-state Kalman filters can be used, the calculation and information requirements are significantly reduced. The control error variance for centralized and decentralized networks, with and without measurement cross-covariance, are compared, and the advantage of centralized fusion in all cases is shown