"Warning: Objects in Driverless Car Sensors May Be Closer Than They Appear"
Researchers at Duke University have demonstrated an attack in which industry-standard autonomous vehicle sensors are tricked into perceiving nearby objects as closer or further than they appear without being detected. Their study suggests that the protection of autonomous attacks from attacks may require adding optical 3D capabilities or the ability to share data with nearby cars. One of the most significant challenges faced by those who develop autonomous driving systems is ensuring protection against attacks. One of the common strategies for ensuring safety is to compare data from separate instruments to make sure their measurements are accurate together. Today's autonomous car companies mostly use locating technology that combines 2D data from cameras and 3D data from LIDAR, which is laser-based radar. Although this combination has been proven to be robust against various attacks that attempt to fool the visual system into seeing the environment incorrectly, the researchers have discovered limitations to the system. Their research shows how adding a few data points in the 3D point cloud ahead or behind an object's actual placement can cause these systems to make incorrect decisions. The researchers' attack strategy involves shooting a laser gun into a car's LIDAR sensor to add false data points to its perception. If the data points are out of place with what a car's camera is perceiving, previous studies showed that the system could recognize the attack. However, according to the new study, the system can be fooled by carefully placing 3D LIDAR points within a certain area of a camera's 2D field of view. The team says this attack can trick adaptive cruise control into thinking a vehicle is slowing down or speeding up. This article continues to discuss the attack demonstrated by the researchers that can deceive industry-standard dual-camera LIDAR sensors on autonomous vehicles and suggested strategies for improving the security of self-driving cars against such attacks.
Duke University reports "Warning: Objects in Driverless Car Sensors May Be Closer Than They Appear"