Autonomous driving alarm: the car mistakes red for green

ROME – The problem of possible attacks on self-driving car systems comes back to the fore with a study by researchers from Zhejiang University in Hangzhou, China. In the course of the tests carried out by Chen Yan and his colleagues, it emerged, in fact, that driverless cars can be induced to interpret the red light emitted by traffic lights as green when they are red. By aiming a laser at the cameras used in self-driving cars, the team of researchers found that the guidance system interpreted red light from traffic lights as green in thirty percent of cases. Obviously, the vulnerability found translates into a danger to the safety of autonomous cars which could thus cause serious accidents.

The results of the study would therefore tend to confirm the concerns that have been circulating for some time about the possible exposure of self-driving vehicles to possible attacks by malicious actors. In the course of its research, the team of Chinese scholars aimed a laser at the sensors of five camera models mounted on driverless cars, using two simple open source software packages for the interpretation of the images acquired by the camera on duty.

According to the study authors, it is desirable to redesign cameras for self-driving vehicles so that they scan rows of pixels in random order, rather than one row at a time, thus making it more difficult for an attacker to introduce color interference. . Richard Smith of the UK’s De Montfort University is also of the same opinion who supports the usefulness of research such as this, as they highlight the need for specific legislation for driverless cars, which are not yet supervised by a global regulatory body. . “Autonomous vehicles will have to conform to strict guidelines like airlines,” said Smith. “If that doesn’t happen, there will be serious problems.

This is why studies like this are so important, because they help highlight potential problems that car manufacturers need to think about right now. ” It should be noted that the experiment required a lot of work to deceive a single sensor and it is hoped that driverless cars will have systems based on multiple cameras, even if, according to Smith, a determined attacker with sufficient resources “will always be able to create damage “.