Sensor Fusion Architectures for Autonomous UAV Navigation

This paper presents an innovative method to estimate the performances of different candidate sensor fusion architecture for the navigation of autonomous Unmanned Aerial Veichles. Indeed, the identical navigation system of manned aircrafts can not be adopted onboard unmanned ones because they require increased levels of autonomy that compensate the absence of the human pilot. At present, no single sensor exist that can perform standalone navigation for UAVs. For this reason, proposed solutions involves the adoption of multiple sensor configuration controlled by a data fusion logic. In typical configurations, inertial sensors are used as main source of navigation reference and other sensor such as GPS, magnetometers, air data sensors, and radar or laser altimeters are used as aiding sensors. Some configurations include all the reported systems, even if they are not always synchronously used but only in critical mission phases such as landing or terrain following. At project level, one important task is to estimate the accuracy of the navigation system architecture composed by the sensors and the integration logic. Kalman filters are often used as sensor fusion algorithms. No analytical method is known to evaluate their performances. Indeed, two main numerical methods are used such as Covariance Propagation and Montecarlo Simulation. Anyway both methods have limitations. Covariance Propagation often outputs “best case” performance instead of typical ones. Montecarlo Analysis requires an heavy processing activity to perform several simulation runs. The proposed method determines the Covariance Matrix of the error state in the Kalman filter in steady-state conditions. This method requires a finite number of computations and guarantees implicit convergence to steady-state solution. Equations governing the method are fully discussed and the results are compared to the ones of the other available techniques.