Location-Based Sensor Fusion for UAS Urban Navigation.

Abstract

For unmanned aircraft systems (UAS) to effectively conduct missions in urban environments, a multi-sensor navigation scheme must be developed that can operate in areas with degraded Global Positioning System (GPS) signals. This thesis proposes a sensor fusion plug and play capability for UAS navigation in urban environments to test combinations of sensors. Measurements are fused using both the Extended Kalman Filter (EKF) and Ensemble Kalman Filter (EnKF), a type of Particle Filter. A Long Term Evolution (LTE) transceiver and computer vision sensor each augment the traditional GPS receiver, inertial sensors, and air data system. Availability and accuracy information for each sensor is extracted from the literature. LTE positioning is motivated by a perpetually expanding network that can provide persistent measurements in the urban environment. A location-based logic model is proposed to predict sensor availability and accuracy for a given type of urban environment based on a map database as well as real-time sensor inputs and filter outputs. The simulation is executed in MATLAB where the vehicle dynamics, environment, sensors, and filters are user-customizable. Results indicate that UAS horizontal position accuracy is most dependent on availability of high sampling rate position measurements along with GPS measurement availability. Since the simulation is able to accept LTE sensor specifications, it will be able to show how the UAS position accuracy can be improved in the future with this persistent measurement, even though the accuracy is not improved using current LTE state-of-the-art. In the unmatched true propagation and filter dynamics model scenario, filter tuning proves to be difficult as GPS availability varies from urban canyon to urban canyon. The main contribution of this thesis is the generation of accuracy data for different sensor suites in both a homogeneous urban environment (solid walls) using matched dynamics models and a heterogeneous urban environment layout using unmatched models that necessitate filter tuning. Future work should explore the use of downward facing VISION sensors and LiDAR, integrate real-time map information into sensor availability and measurement weighting decisions, including the use of LTE for approximate localization, and more finely represent expected measurement accuracies in the GPS and LTE networks.