A team of researchers from well-known universities, including SUNY Buffalo, Iowa State, UNC Charlotte, and Purdue, managed to transform an autonomous car controlled by the Apollo platform from Chinese Internet giant Baidu into a potential danger to other road users by tricking its multi-sensing system using improvised methods.
“Numerous experiments on actual autonomous vehicles demonstrated that the proposed attack can consistently conceal the target vehicle from the perception system of an unmanned vehicle using just two small hostile objects,” clarified by researchers. Their findings were presented at the 30th annual International Conference on Mobile Computing and Networking. (source)
While previous researchers focused on software vulnerabilities in autonomous vehicles, this team broadened the manipulation techniques and deceived systems that rely on Lidar, cameras, and radar concurrently.
The new attack involves reflecting millimeter waves from a smooth metal surface to manipulate unmanned systems. The researchers utilized “inexpensive” and “readily manufactured” items such as cardboard, metal foil, and multi-colored images.
“By positioning a smooth metal surface between the radar and the target vehicle at a specific angle, it is possible to deflect the emitted millimeter waves from the radar receiver, resulting in a reduction in the energy of echo signals from the vehicle,” explained the study’s authors. “Once the energy drops below a certain threshold, the target vehicle becomes undetectable to radar perception.”
The colored markings on the target car distort the pixel values of the input image and affect the Apollo camera’s perception, while reflections challenge the operation of Lidar lasers. Consequently, all three sensory systems start to malfunction, causing the car to disappear from radar detection.
Researchers propose that such an attack could be carried out using drones that “conceal” a secondary vehicle from an unmanned car, projecting or carrying a hostile object:
With only a pair of drones controlled by attackers, this system could trigger a significant traffic accident and swiftly disappear in an unknown direction without a trace.
Alternatively, in the absence of drones, the system could be mounted on the back of a vehicle and disguised under an advertising banner, as depicted in the image below:
Although the Baidu Apollo platform was utilized in this instance, the tactics employed could theoretically be adapted to other detection systems employed in autonomous vehicles.
In China, such attacks could have severe repercussions since unmanned taxis have been actively in use since November 2021, operating in