Demo: Attacking Multi-Sensor Fusion based Localization in High-Level Autonomous Driving

Infrastructure-Enabled Gps Spoofing Detection and Correction

“…Traditional GPS spoofing detection methods rely on hardware for monitoring signal power or arrival angle (Shen et al, 2020). Due to their reliance on special hardware, these strategies have limited application scenarios like airborne GPS receivers.…”

“…Consequently, isolating the attacked sensor is not trivial. Second, in the presence of detection latency, the data fusion framework would have been partially compromised before noticing an attack and isolating the attacked sensor (Shen et al, 2020). Previous studies only emphasize isolating the attacked sensor but lack discussions on correcting the compromised data fusion framework.…”

“…For example, a recently stealthy GPS spoofing is proposed to gradually drift the true vehicle position according to its kinematic model (Wang et al, 2021b). Shen et al (2020) designs a stealthy GPS spoofing attack (named FusionRipper) that could fail production-grade autonomous driving systems (e.g., Baidu's Apollo system) with over 90% success rate. FusionRipper targets the predominantly adopted Multi-Sensor Fusion (MSF) algorithms in practice for localization and injects falsified GPS measurements by exploiting the vulnerability of MSF.…”

“…Vehicle (global) location at time 𝑘 is represented by the KF's state 𝒙 ̂𝑘 and state uncertainty with a covariance matrix 𝑷 ̂𝑘. Due to the non-linearity of the vehicle motion model (Equation ( 1)), we adopt an Extended Kalman Filter (EKF) proposed in Shen et al (2020), which is a typical localization algorithm to fuse IMU and GPS measurements.…”

“…For supporting high-level driving automation and safety, the localization module needs to be robust in various driving scenarios and thus demands advanced sensors and algorithms. Modern localization modules are predominantly based on datafusion algorithms that combine position information from multiple sensors, including, for example, Global Positioning System (GPS), Inertial Measurement Unit (IMU), Light Detection and Ranging (LiDAR) and camera (Shen et al, 2020). However, recent studies suggest that sensors on vehicles are vulnerable to malicious attacks (Petit and Shladover, 2015).…”