Enhancing Autonomous Vehicle Security: Federated Learning for Detecting GPS Spoofing Attack

Autonomous vehicles (AVs) are poised to transform modern transportation, providing superior traffic management and improved user experiences. However, there exists a considerable risk to the acquisition of Position, Velocity and Time (PVT) in AVs, since the acquisition of PVT is vulnerable to Global Positioning System (GPS) spoofing attacks that could redirect the AV to wrong paths or lead to security threats. To address these issues, we propose a novel approach for detecting GPS spoofing attacks in AVs using Federated Learning (FL) with trajectories obtained from the Car Learning to Act (CARLA) simulator. Each vehicle autonomously performs localization using sensor data that includes yaw rate, steering angle, as well as wheel speed. The obtained localized coordinates (authentic and spoofed) are utilized to compute weights. These weights are aggregated at the Roadside Unit (RSU) and shared with the global model utilizing Support Vector Machines (SVM) for classification. The global model updates local models through FL, ensuring data privacy and collaborative learning. The experimental results show that the proposed model achieves 99% accuracy, 98% F1 score, and the AUC-ROC of 99% outperforming traditional machine learning methods including the K-Nearest Neighbors (KNN) and Random Forest (RF). The results demonstrate the practicality of using FL to improve the security of AVs against GPS spoofing attacks with limited data sharing, thereby offering a potential approach for real-world applications.