Detecting CAN Bus Attacks in Autonomous Vehicles Using Deep Learning

Today's autonomous vehicles are equipped with many sensors and devices that are constantly sending and receiving information to keep the car function smoothly. The most common way for these components to communicate with each other is through a system called a Controller Area Network (CAN). The design of the CAN system does not include security mechanisms which makes it easier to attack the system by sending unauthorized messages. Such cyber attacks allow the attacker to take the control of car, thus endangering the passengers and the pedestrians. This study presents a new method that uses deep learning to detect and analyze attacks on the CAN bus. It can identify patterns in the incoming traffic on the CAN bus that indicate attacks, thus increasing the security of modern vehicles against cyber threats. The proposed system is trained and validated on an online available dataset to detect various attacks against the CAN bus in autonomous vehicles and it achieved an average accuracy score of 99.5%, thus ensuring the safety and reliability of CAN. The results show the efficacy and robustness of the proposed system in the detection and classification of well-known cyber-attacks in autonomous vehicles.