Enhanced Bayesian State Space Estimation for a GNSS/INS Tightly-Coupled Integration in Harsh Environment: An Experimental Study

Abstract

Several navigation filters have been developed since the early steps of Global Navigation Satellite Systems (GNSS) to provide high-accuracy Positioning, Navigation and Timing (PNT), and many solutions are available in the literature to support a plethora of applications. In the context of vehicular navigation and positioning, advanced state estimation and sensors fusion techniques cannot cope by themselves with strong multipath effects in dense urban areas. Therefore, these solutions require more robust approaches typically involving an additional processing effort, especially in low-cost Inertial Navigation System (INS)/GNSS Tightly-Coupled (TC) integration scheme. This work analyzes state-of-the-art covariance matrix estimation methods and proposes an INS-based pre-processing stage to mitigate the impact of undesired, multipath-related bias injections without inhibiting the Inertial Navigation System (INS)/GNSS integration. The proposed adaptive solution improves the overall stability and estimation accuracy of a set of Bayesian filters, i.e., Extended Kalman Filter (EKF), Unscented Kalman Filter (UKF), and Unscented Particle Filter (UPF). Results are presented about a real dataset in which an advanced, state-of-the-art Adaptive UPF (AUPF) TC scheme, applied to a low-cost integrated setup, occasionally failed to track the navigation solution due to poorly conditioned GNSS measurements.