Navigation Context Adaptive Fault Detection and Exclusion Strategy based On Deep Learning & Information Theory: Application to a GNSS/IMU Integration
Nesrine Harbaoui, Nourdine Aït Tmazirte, Khoder Makkawi, Maan El Badaoui El Najjar · Proceedings of the Satellite Division's International Technical Meeting (Online)/Proceedings of the Satellite Division's International Technical Meeting (CD-ROM) · 2021
The growing demand for a reliable localization function in safety-relevant autonomous applications has prompted the navigation community to take a close interest in Fault Detection algorithms. Particularly powerful, these methods have as common limitation, the need to define a priori probabilities of fault(s) presence or absence in order to set a detection threshold. In the case of systems insensitive to external phenomena, these probabilities can be obtained although with difficulty. However, in a multi-sensor localization system using, among other, Global Navigation Satellites System (GNSS), this task can prove to be delicate. Indeed, these global navigation systems suffer from perturbations induced by the local environment (buildings in urban canyons, foliage in forests, intentional or nonintentional interferences...) causing local feared events such as multipath, Non-Line-Of-Sight (NLOS) or GNSS outage. The a priori probabilities to face this type of event, and therefore to observe one or more inconsistent measures, are very unpredictable. This makes the task of setting a threshold difficult. In this study, we investigate the combined use of 1) GNSS expertise: allowing to identify the different parameters, making it possible to monitor the state of health of the measurements, and therefore of the global system, and 2) the implementation of a deep learning scheme, as a decision-making support, considering these inputs, and providing the adequate a priori probability of fault presence.