Securing Self-Driving Vehicles with Artificial Intelligence
A new system could better protect self-driving vehicles from hackers and bad signals by using artificial intelligence (AI) to secure wireless communications. With human lives on the line, it is crucial self-driving vehicles are secured in the design stages rather than in afterthought spot patches. With countless data messages coming through each vehicle sensor per second, systems must ensure each signal received is legitimate. This leads to unsustainably high computational overhead, putting connected vehicle systems at risk.
A self-driving vehicle communicates through a complex network of sensors. When on the road with other self-driving and connected cars, it must sift through the hundreds of messages generated each second by its own sensors, as well as the sensors of other self-driving vehicles. This can cause confusion and lead to a car getting the wrong directives. Additionally, hackers can create dummy transmitters to create false roadblocks, leaving passengers at the hackers’ mercy.
To address this issue, researchers at the University of Massachusetts propose a method employing weighted sensor readings along with artificial intelligence to protect passengers in self-driving vehicles from erroneous or even malicious messages. The researchers’ system is cheaper and requires less computation than other message verification systems.
Connected vehicles currently connect to each other wirelessly with the Vehicular Ad-hoc Network (VANET). VANET is a wireless communication system that uses mobile cars as communication nodes to determine positioning and movement for connected vehicles.
Wireless networks, like VANET, validate every message circulating in their network without assessing the source’s quality or the content’s accuracy. As a result, VANET is inherently vulnerable to false messages being spread in the network by message spoofing and denial-of-service (DoS) attacks.
“The security of connected vehicles plays a pivotal role as they can be used to prevent life threatening situations,” said Prinkle Sharma, doctoral researcher of computer engineering at the University of Massachusetts. “Falsification of meter readings, disablement of brake function and other unauthorized controls by spoofed messages injected into the Vehicular Ad-hoc Network can cause injury to passengers and damage to the vehicles. Countermeasures must be considered at the design stage, as opposed to creating solutions after an incident.”
To solve the message validation vulnerability of VANET, the researchers created a system that augments message authentication with artificial intelligence using particle filters. The artificial intelligence analyzes the data provided by the sensors and identifies irregularities to be verified further.
The AI checks signals from new vehicles and learns to accept signals from reliable transmitters moving forward. Another way this system protects passengers is by preventing spoofed messages. Algorithms embedded in the system are used for future position prediction. If another vehicle is not behaving as predicted by the algorithms, then a response is triggered to verify the source of the recorded position.
Validating select, rather than all, messages protects VANET from DoS attacks without losing the effectiveness of faulty message detection. The researchers found their system can save up to 85 percent on computational overhead using signal processing filters. The efficiency of artificial intelligence allowed for better verification of relevant messages.
Due to the massive amount of data transmission, the proposed system is not ready for deployment in self-driving vehicle fleets. The number of spoofed messages missed by the AI in this testing was about 11%, which is too high a rate when human lives are on the line. The researchers are now conducting additional testing and making adjustments to lower this rate to a more acceptable number. By performing intensive training using deep learning techniques, they hope to enhance the system’s scalability, security, and decision-making capability inside the network.
Further research and testing is needed, but a safe, self-driving vehicle future is one step closer to becoming a reality.
For more information on self-driving vehicles, visit the IEEE Xplore Digital Library.