LAUSR

Tight integration algorithms fusing Global Navigation Satellite System (GNSS) and Inertial Navigation System (INS) have become popular in many high-accuracy positioning and navigation applications. Despite their reliability, common integration architectures can still run into accuracy drops under challenging navigation settings. The growing computational power of low-cost, embedded systems has allowed for the exploitation of several advanced Bayesian state estimation algorithms, such as the Particle Filter (PF) and its hybrid variants, e.g. Unscented Particle Filter (UPF). Although sophisticated, these architectures are not immune from multipath scattering and Non-Line-of-Sight (NLOS) signal receptions, which frequently corrupt satellite measurements and jeopardise GNSS/INS solutions. Hence, a certain level of modelling adaptivity should be granted to avoid severe drifts in the estimated states. Given these premises, the paper presents a novel Adaptive Unscented Particle Filter (AUPF) architecture leveraging two cascading stages to cope with disruptive, biased GNSS input observables in harsh conditions. A INS-based signal processing block is implemented upstream of a Redundant Measurement Noise Covariance Estimation (RMNCE) stage to strengthen the adaptation of observables’ statistics and improve the state estimation. An experimental assessment is provided for the proposed robust AUPF that demonstrates a 10 % average reduction of the horizontal position error above the 75-th percentile. In addition, a comparative analysis both with previous adaptive architectures and a plain UPF is carried out to highlight the improved performance of the proposed methodology.