Hybrid multi-sensor navigation system with uncertainty correction for GNSS-denied environments

Acknowledging the vulnerabilities of the Global Navigation Satellite Systems (GNSS) to various interferences, this research investigates alternative navigation solutions, essential for overcoming challenges where GNSS quality is compromised. The study explores a multi-sensor solution, suitable for operation in complex scenarios including degraded environmental conditions. To mitigate the inherent drifting behavior of a widely used alternative navigation information source referred to as the Inertial Navigation System (INS), fusion with a camera and a barometer is adopted within the federated multi-sensor architecture. This approach utilizes an error detection mechanism based on analysis of residual test statistics for Extended Kalman Filter (EKF)-based local filters. To enhance the robustness of the system, a Bidirectional Long Short-Term Memory (BiLSTM) model is implemented for error correction of the filter measurements, integrated before fusion in the master filter. Validation tests in a simulated urban environment using various trajectories and environmental conditions reveal that the proposed mechanism provides a viable alternative to a GNSS-based system for positioning. The performance is compared with the state-of-the-art learning-based multi-sensor navigation system by testing on similar datasets. Comparative results indicate significant improvements in positioning with error correction yielding enhancements of 34%, 44%, and 20% in rainy, snowy and foggy conditions, respectively.