Self-driving cars are probably not as smart as they are thought to be. Even the smallest specks of paint could disrupt the autopilot – with fatal consequences. Hacker attacks are also easily possible, researchers speculate based on a new study.
Self-driving cars should be able to react faster than humans. However, there have already been headlines about self-driving cars in which the technology didn’t work so perfectly after all, which is why some of the accidents were fatal. Researchers at the max planck Institute for intelligent systems (MPI-IS) in tubingen have now discovered that even the smallest specks of color and patterns could confuse the autopilot to such an extent that the vehicles could potentially become a real safety risk. The scientists’ findings are still only conjecture, because individual car manufacturers do not publish exactly how their self-driving cars work. The result of the work is nevertheless alarming.
How autonomous vehicles work
Cameras attached to self-driving cars provide data on the surrounding environment. A system evaluates and interprets them. In simple terms, this is how a self-driving car should work. How this interaction works in detail is a well-kept secret of the respective manufacturers.
Movement in a scene is misinterpreted
Researchers have already been able to prove several times in the past that objects such as stop signs are misclassified during data transmission. The tubingen researchers led by anurag ranjan have now been able to prove for the first time that the system misinterprets the movement in a scene in the event of a malfunction. To cause such a disturbance, even a small spot the size of less than one percent of the entire image is sufficient. The consequences are fatal: the on-board computer assumes false facts. According to his calculation, the objects move in the opposite direction than in reality. This happens even if the color pattern does not move. Basically, the bigger the stain, the more devastating the effects.
Hackers also have an easy game
Hackers could also easily manipulate cars because of this susceptibility to interference, as the tubingen researchers were able to prove.
In plain language, this means that a color pattern on a t-shirt, as a rear window sticker or as an emblem on a shopping bag could pose a problem for self-driving cars.
image rights: BR
Autonomous cars must know all by themselves whether they are encountering a pedestrian or whether there is just a trash can at the same spot. Researchers have now been able to confuse the software of autonomous cars – with a spot of color.
Researchers’ goal: inform carmakers about risks
The researchers have a clear goal:
They have therefore informed companies such as Tesla, Porsche, Daimler, BMW and Bosch about the possible risk.
Autonomous vehicles: "still a long way from series production"
But there is one consolation: the risk of series-production vehicles currently available on the market being affected is low, according to the press release from MPI-IS in Tubingen.
© MPI for intelligent systems / W. Scheible
Image rights: MPI for Intelligent Systems / W. Scheible
With this color pattern, researchers at the max planck institute in tubingen were able to mess up a self-driving vehicle.