Poster 1: Security Analysis of Multi-Sensor Fusion based Localization in Autonomous Vehicles
(Room Palms Ballroom)
16 Jul 19
5:30 PM
-
7:00 PM
Tracks:
Vehicle Automation Technology
Precise and robust localization is highly critical for making correct and safe driving decisions in Autonomous Vehicles (AVs). To achieve this goal, they are usually equipped with redundant and complementary sensors, e.g., LiDAR, GPS, and IMU, and use Multi-Sensor Fusion (MSF) algorithms to combine the observations. However, it is largely unclear how robust these MSF algorithms are in the presence of practical sensor attacks such as GPS spoofing. This work presents the first security analysis of representative MSF designs in AV systems under the GPS spoofing attack. To perform the analysis, a synthetic driving scenario is designed to overcome the challenges in the handling of sensor noises and the lack of ground truth. Next, the authors formulate the analysis task as an optimization problem to understand the upper bound of attack capability, and then leverage the analysis insights to identify effective spoofing strategies. The initial results show that a well-designed spoofing strategy is able to deviate the localization estimation of a representative MSF implementation by 2 meters in as short as 10 seconds.