"Autonomous Vehicles Can Be Fooled to 'See' Nonexistent Obstacles"

Autonomous vehicles see the world through the use of multiple sensors. Most autonomous vehicle systems use cameras, radar sensors, and LiDAR (light detection and ranging) sensors. The data collected through cameras and sensors are combined by an onboard computer to get a complete view of what is surrounding the car. Although the use of multiple sensor systems significantly increases the safety of vehicles, they are still at risk of facing attacks. Researchers have shown that the machine learning algorithms used in camera-based perception systems can be fooled by placing stickers on traffic signs. Another study conducted by the RobustNet Research Group at the University of Michigan demonstrated the possibility of tricking the LiDAR-based perception system into seeing a nonexistent obstacle, such as another car, by spoofing the LiDAR sensor signals. The researchers performed two different LiDAR spoofing attacks on a widely-used autonomous driving system, called Baidu Apollo. This article continues to discuss the components of a LiDAR-based perception system and the demonstration of LiDAR spoofing attacks designed by researchers. 

GCN reports "Autonomous Vehicles Can Be Fooled to 'See' Nonexistent Obstacles"