Sensor fusion algorithms to improve safety and real-time capability of vehicular navigation systems

Abstract

Over the last ten years, much work has gone into the development of autonomous vehicles, and the DARPA Urban Challenge in 2007 showed that autonomous vehicles will be developed in the near future. Currently, research on commercial autonomous vehicles with various low-cost sensor is accelerating, and sensor fusion techniques for autonomous vehicles are also gaining attention. Factors that degrade the the performance of the conventional sensor fusion techniques include sensor failure, overly complex computational requirements, and external attacks (i.e., jamming and spoofing). This dissertation introduces a two stage federated Kalman filter (TS-FKF) and deterministic compressed sensing based reduced order Kalman filters (DCS-ROKF and DCS-ROEKF) which correspondingly improve the safety and real-time capability of navigation systems. The TS-FKF is demonstrated with information sharing and slip and slide detection algorithms for high-speed trains, and the DCS-ROKF and DCS-ROEKF are demonstrated for autonomous vehicles. The simulation results of tests of the TS-FKF and DCS based filters (DCS-ROKF and DCS-ROEKF) show that they successfully improve the safety and real-time capability of their respective navigation systems. Recently, cooperative navigation systems utilizing vehicle-to-vehicle (V2V) and vehicle-to-infra (V2I) communication techniques have been developed to surpass the limits of autonomous vehicles. Cooperative navigation systems are capable of better navigation performance than multi-sensor based navigation systems, as they integrate information from their own sensors, connected neighboring vehicles, and the connected infra. However, the performance of these cooperative navigation systems is degraded in urban environments due to non-line-of-sight (NLOS) delays or malicious attacks. To develop robust cooperative navigation systems capable of functioning in urban environments, secure urban cooperative positioning techniques (SUCPTs) are introduced in this dissertation. The simulation results of SUCPTs demonstrate that they are robust against NLOS delays and malicious attacks through their capabilities to detect NLOS satellites, NLOS vehicles, and malicious vehicles. Robust navigation systems can be implemented with the techniques in this dissertation, and the computational complexity of these systems can be reduced. Therefore, this dissertation contributes to the development of autonomous and cooperative navigation systems for various vehicles such as cars, trains, and airplanes.