1. Home
  2. RIP
  3. View Record
https://nap.nationalacademies.org/catalog/27432/critical-issues-in-transportation-for-2024-and-beyond

Resilience and Validation of GNSS PNT Solutions

The study team will take a proactive approach to assess vulnerabilities in specific case studies. They will also develop new PNT sensing techniques for HATS that are both robust in the face of unusual natural or accidental events and secure against deliberate attack. The team's insights will then be generalized, leading to the development of broad PNT protocols, procedures, and recommendations for standards. In addition, to investigate possible mitigation strategies in the event of a PNT lapse, much insight can be gained by creating a set of hypothetical scenarios in a traffic assignment model that simulate and characterize the effects of possible attacks. The measured severity of problems in each scenario can then inform an estimation of risk and the best types of mitigation. For example, scenario results can highlight the positive and negative aspects of broadly safeguarding selected traffic corridors or urban regions instead of finely limiting possible damage to individual intersections. ++++++++++ (Task 3.3) Low-Cost Ground Vehicle Anti-Spoofing: The study team proposes to develop a technique that exploits road roughness, MEMS IMUs, and carrier phase GNSS to achieve low-cost anti-GNSS-spoofing appropriate for mass-market ground vehicles. The technique will continuously monitor a detection statistic that measures the consistency between the millimeter-precise received GNSS carrier phase measurements and accelerometer measurements. Spoofing is declared when the detection statistic crosses a constant-false-alarm-rate threshold. The results will be a low-cost GNSS spoofing detection technique that enables a ground vehicle to reliably and promptly detect code-phase-aligned GNSS spoofing despite multipath, signal blockage, and other urban effects. ++++++++++ (Task 3.5) Physics-Based Anomaly Detection: The study team proposes to evaluate existing cyber-layer defense PBAD strategies and also develop new ones for PNT threats to HAVs by (1) identifying promising data sources in HAVs (e.g., sensing data and related actuators) that can potentially correlate with the PNT inputs, (2) understanding their correlations and constructing physical invariants, (3) designing online anomaly detection algorithms corresponding to the identified physical invariants, and (4) evaluating the performance in real-world HAVs. The team will quantify the safety and mobility consequences (e.g., increased crashing rate and string stability) from real-world or simulation-based experiments for each class of PNT threats and design and implementation of the mitigation methods leveraging the PBAD method. ++++++++++ (Task 3.6) Machine Learning PNT Model Security: Some PNT sensors used in HAVs (e.g., lidars and cameras) predominately rely on deep neural networks (DNNs). The study team proposes to perform risk analysis and identify mitigation methods for security issues at the DNN model level. Specifically, in the recent race between DNN attacks and defenses, numerous defense/mitigation techniques have been proposed, e.g., model ensembling, randomization, image transformation, and adversarial training. The team plans to first explore domain-specific adaptions of existing defense strategies in other application domains to 10the DNN models used in PNT algorithms and then leverage the insights to design effective solutions specific to PNT threat scenarios. The team will quantify the safety and mobility risks from real-world or simulation-based experiments under deep neural network (DNN) based PNT model attacks, and design and implementation of mitigation methods, such as model architecture changes and adding model input sanitization. ++++++++++ (Task 4.4) Receiver Data Validation by PNT Solutions from Other Sources: Data to create or contribute to an alternative PNT solution that can be used to validate a GPS-based solution may come from different sensors, deployed on different platforms and acquired at different times. Sensors, such as radar, lidar, vision camera, etc., on an HAV provide data in real-time. In contrast, an HD map represents sensor data acquired earlier and structured for some purpose. Connected HAVs may share both their PNT solutions and sensor data. The signals used to observe the environment and then create a PNT solution may come from infrastructure dedicated to positioning, such as GPS or from other sources of natural and man-made ones. Examples of man-made SoPs include cellular and LEO satellites. Natural signals include Earth gravity, magnetic field, and surface-based signals used in terrain-based navigation. Other CARMEN project will provide discussion on many of these alternative PNT solutions. In this work the study team will combine both the theoretical and experimental results into a performance assessment/validation document. Note that the reliability and vulnerability of the various sensor streams may not be considered in the analysis here due to resource limitations.