The Ghost Navigator: Revisiting the Hidden Vulnerability of Localization in Autonomous Driving

Localization is crucial for Autonomous Driving (AD), which serves as a critical foundation impacting the performance of downstream modules. While Multi-Sensor Fusion (MSF) techniques enhance localization accuracy and reliability, the security of fusion-based localization systems has emerged as a major concern. Although existing studies have extensively investigated security aspects of these systems, the impact of vehicle dynamics on the effectiveness of Global Positioning System (GPS) spoofing attacks is persistently overlooked. Bridging this research gap, we propose the Motion-Sensitive Analysis Framework (MSAF), which focuses on analyzing previously underestimated dynamic behaviors of vehicles. Our investigation demonstrates that two dynamic scenarios, acceleration and high-speed cruising, significantly influence the success rates of GPS spoofing attacks. These scenarios, commonly encountered across driving conditions, exhibit heightened vulnerabilities under MSAF analysis. Building on these insights, we design two dynamics-targeted attack strategies and evaluate them across three testbeds: our simulated framework (MSAF_MSF) and two real-world MSFbased autonomous driving systems (Apollo_MSF and Shen-lan_MSF). The results demonstrate a significant attack efficiency improvement by our method: MSAF requires substantially less time to complete attacks compared to the baseline while achieving higher success rates. Code and attack demos are available at https://sites.google.com/view/msaf-attack .