Spoofing attack technology of autonomous vehicle multi-sensor navigation system

Abstract

In recent years, autonomous driving technology has been rapid advancements and has become a critical development in next-generation vehicle technology. Autonomous vehicles (AVs) require centimeter-level positioning accuracy for safe navigation. With the rapid advancement of autonomous driving technology, research on spoofing attacks against the multi-sensor fusion (MSF) systems of AVs has garnered widespread attention. Spoofing attacks refer to the behavior of transmitting false signals to satellite receivers. The navigation systems of AVs typically consist of multiple sensors, including light detection and ranging (LiDAR), inertial navigation systems (INS), and global navigation satellite systems (GNSS). MSF systems can suppress spoofing signals, thereby increasing the difficulty of successfully executing spoofing attacks on AVs. In-depth research on spoofing technology can effectively expose existing vulnerabilities in the field of AVs, which is of great significance for advancing anti-spoofing technology for AV navigation in complex environments. Current research primarily focuses on generating GNSS spoofing signals and designing and defending spoofing algorithms based on GNSS/INS integrated navigation systems. However, prior studies have not provided a detailed analysis of spoofing MSF behavior, such as the development of analytical models and error mechanism analysis. Since the navigation systems of AVs are typically composed of multiple navigation sensors, traditional spoofing methods are easily detected by the system, resulting in failed deception attempts. AVs may operate in diverse geographical and weather environments. Conventional approaches have not fully evaluated the effectiveness of spoofing, and there is an issue of indiscriminately broadcasting spoofing signals. Therefore, in response to these unresolved challenges, this thesis conducts in-depth research. The main research content is as follows: Targeting the issue of traditional error propagation models based on Kalman filters, which are complex in derivation due to differences in sensor update frequencies and numerous inversion operations, and where the mathematical relationship between state errors and system parameters is unclear under spoofing attacks, this study investigates a state error mechanism based on a lightweight information filter model. First, considering the impact of varying sensor update frequencies, an error state Kalman filter analytical model is established. The measurement update processes of GNSS and LiDAR, as well as the INS state recursion process, are derived. The primary factors leading to increased state errors are analyzed. In addition to the initial state covariance matrix and LiDAR uncertainty, the uncertainty of GNSS and the update frequency of measurement sensors are also key factors affecting the spoofing success rate. To avoid the complex inversion operations in multiple LiDAR measurement updates within one GNSS update cycle, an analytical model based on a lightweight information filter is established. Finally, the state error vector update process is transformed into an information vector update process, and inversion operations in the information vector update process are avoided by disregarding the INS recursion update process. The state error propagation model has a clearer mathematical expression, intuitively reflecting the mathematical relationship between state errors and system parameters, thereby providing a theoretical foundation for research on spoofing technology in navigation systems across various environments. Addressing the challenge that traditional spoofing methods struggle to adapt spoofing parameters according to spoofing effects, resulting in low spoofing success rates, a covert spoofing method based on a fuzzy inference model is proposed. This method involves monitoring the target AV, calculating real-time position error feedback adjustment factors, constructing fuzzy knowledge bases and fuzzy rules, and dynamically adjusting spoofing parameters using the multi-Zadeh method to improve spoofing success rates. By comparing position error feedback adjustment factors before and after real-time adjustments, the method determines whether the spoofing process has triggered the take-over effect. If the take-over effect is triggered, constraints on the maximum values of spoofing parameters are enforced. Real-world data test results demonstrate that the proposed method achieves a spoofing success rate 5% higher than traditional methods in typical test scenarios. Regarding the issue that traditional spoofing technologies do not evaluate the effectiveness in complex geographical and adverse weather conditions and blindly launch spoofing attacks, this study investigates a spoofing effectiveness evaluation method based on sensor uncertainty estimation. For complex geographical scenarios, a three-dimensional building model of the target area is constructed. A sky visibility mask is generated based on the maximum elevation angle edge of the building model, and sky visibility and the number of visible satellites are estimated. A kernel partial least squares nonlinear regression model is established to estimate GNSS uncertainty, analyzing the relationship between sky visibility and spoofing success rates. For various weather scenarios, the impact of weather on LiDAR performance is evaluated through meteorological pulse response functions at different weather levels, including rain, snow, and fog. A LiDAR uncertainty estimation method based on a B-spline regression model is developed, and the relationship between different weather levels and spoofing success rates is quantitatively analyzed. The results indicate that the proposed method can evaluate whether the AV’s environment is conducive to spoofing. Real-world data test results show that in different geographical scenarios, when the proposed method determines the scenario is easy, the spoofing success rate exceeds 70%, outperforming traditional methods. In various weather scenarios, when the proposed method determines that the scenario is easy, the spoofing success rate exceeds 57%. This study provides a theoretical foundation for designing defense algorithms in the event of potential malicious spoofing attacks, facilitating further research into GNSS anti-spoofing algorithms. Thereby, it enables vehicles to implement proactive measures, such as activating emergency plans, slowing down, or even stopping in the event of safety risks. Immediate restoration of MSF system performance ensures vehicle safety and reduces the risk of catastrophic traffic accidents. Ultimately, this study ensures the safety of the MSF system in various scenarios.