Final Presentation available for NAVISP EL1-087 bis "Verifiable AI/ML techniques for PNT applications"

Last Updated: 12/06/2026 06:51     Created at: 12/06/2026 06:45

On the 10th of June 2026, Fondazione Bruno Kessler presented, with their subcontractor GMV UK, the results of NAVISP-EL1-087 bis (VAIPOSA), which focused on innovative methodologies to design Positioning, Navigation, and Timing (PNT) components using machine learning (ML), while ensuring safety and resilience.

VAIPOSA combined a classical GNSS-based engine with an AI-based visual odometry engine, then supervises them through a “safety cage” architecture that detected faults, evaluates confidence, and dynamically adjusted the contribution of each engine.

The safety cage used both qualitative cut-off rules and quantitative weighting to manage anomalies such as poor luminosity, frozen images, camera failures, and GNSS degradation. 

Results showed that the fused approach improves navigation robustness and accuracy, especially in faulty conditions.

The activity was fully funded by ESA NAVISP Element 1, whose objective is to innovate PNT technologies. The slides can be accessed here.