Black Hat | The Key to Remote Vehicle Control: Autonomous Driving Domain Controller @BlackHatOfficialYT | Uploaded 1 month ago | Updated 13 hours ago
Nowadays, cars are getting smarter, with many featuring advanced driver assistance systems, especially new energy vehicles.
Among these developments, the Autonomous Driving Domain Controller is a crucial part. Many smart cars use AI chips like Nvidia Orin and Mobileye EQ5. They employ high-precision GPS, multiple cameras, LiDAR, and other sensors to sense and locate their surroundings, making decisions using algorithms. They control the car's movements, like braking and steering, by sending and receiving CAN messages.
However, the security risks of these controllers seem to have been overlooked. Research often focuses on in-car entertainment systems, TBoxes, MCUs, etc., rather than the safety of these controllers. If hackers gain control of the autonomous driving domain controller, they could easily manipulate the car's acceleration, steering, and more through API interfaces, posing significant dangers.
In our presentation, we'll thoroughly explain the hardware structure and testing methods of the autonomous driving domain controller. We'll cover common architectures like Mobileye, Nvidia, TI, and Horizon. We'll discuss the software and hardware security risks present in these devices. We'll show how to extract AI recognition models and conduct security tests on them. Additionally, we'll demonstrate, how to gain control of a vehicle's autonomous driving domain controller and control the entire vehicle.
We aim to raise awareness about the critical security issues associated with autonomous driving domain controllers.
By:
Shupeng Gao | Senior Security Researcher, Baidu
Yingtao Zeng | Senior Security Researcher, Baidu
Jie Gao | Senior Security Researcher, Baidu
Yimi Hu | Senior Security Researcher, Baidu
Full Abstract & Presentation Materials:
blackhat.com/asia-24/briefings/schedule/#the-key-to-remote-vehicle-control-autonomous-driving-domain-controller-38089
Nowadays, cars are getting smarter, with many featuring advanced driver assistance systems, especially new energy vehicles.
Among these developments, the Autonomous Driving Domain Controller is a crucial part. Many smart cars use AI chips like Nvidia Orin and Mobileye EQ5. They employ high-precision GPS, multiple cameras, LiDAR, and other sensors to sense and locate their surroundings, making decisions using algorithms. They control the car's movements, like braking and steering, by sending and receiving CAN messages.
However, the security risks of these controllers seem to have been overlooked. Research often focuses on in-car entertainment systems, TBoxes, MCUs, etc., rather than the safety of these controllers. If hackers gain control of the autonomous driving domain controller, they could easily manipulate the car's acceleration, steering, and more through API interfaces, posing significant dangers.
In our presentation, we'll thoroughly explain the hardware structure and testing methods of the autonomous driving domain controller. We'll cover common architectures like Mobileye, Nvidia, TI, and Horizon. We'll discuss the software and hardware security risks present in these devices. We'll show how to extract AI recognition models and conduct security tests on them. Additionally, we'll demonstrate, how to gain control of a vehicle's autonomous driving domain controller and control the entire vehicle.
We aim to raise awareness about the critical security issues associated with autonomous driving domain controllers.
By:
Shupeng Gao | Senior Security Researcher, Baidu
Yingtao Zeng | Senior Security Researcher, Baidu
Jie Gao | Senior Security Researcher, Baidu
Yimi Hu | Senior Security Researcher, Baidu
Full Abstract & Presentation Materials:
blackhat.com/asia-24/briefings/schedule/#the-key-to-remote-vehicle-control-autonomous-driving-domain-controller-38089
