Previous Abstract Return to Session A1 Next Abstract

Session A1: Augmentation Services, Integrity, and Authentication 1

Enhancing GNSS Pedestrian Navigation Reliability through Artificial Intelligence-driven Integrity Monitoring in Complex Environments
Ziyou Li, Ni Zhu, Valérie Renaudin, University Gustave Eiffel, AME-GEOLOC
Date/Time: Wednesday, Sep. 18, 10:40 a.m.

Best Presentation

The signal propagation channel can hardly be accurately modeled in complex environments such as deep urban and light indoors where the GNSS signal is severely degraded. Therefore, traditional statistical-based integrity monitoring (IM) algorithms face severe challenges. The traditional Fault Detection Exclusion (FDE) may fail due to insufficient redundancy, and the Protection Level (PL) is usually overestimated, thus increasing the system unavailability rate. This paper presents a novel Artificial Intelligence (AI)-based IM system, to provide accurate and reliable positioning solutions for pedestrian navigation in complex environments such as deep urban and light indoors. The proposed system first implements the previously proposed Machine Learning (ML)-based FDE algorithm, i.e., ``LIGHT (Light Indoor GNSS macHine-learning-based Time difference carrier phase)", and calculates positioning solution (velocity and heading) with the Time Difference Carrier Phase (TDCP) algorithm. After that, an AI model with Long Short-Time Memory (LSTM) is trained to estimate the uncertainty of velocity and heading calculated by the LIGHT-TDCP to further enhance the reliability of the positioning solution. Two pedestrian datasets in deep urban and light indoor environments with a total walking distance of 1.8 km are tested. Compared with the traditional PL calculation approach, the results show that the proposed approach decreases the system unavailability rate by 57 and 4.7 times for urban and light indoors respectively, and the Misleading Information (MI) rate achieves the level of $10^{-3}$, which meets the standard set by existing pedestrian navigation research.



Previous Abstract Return to Session A1 Next Abstract