Tesla, which disbanded its public relations team in 2021, did not respond to WIRED’s request for comment. The camera systems the researchers used in their tests were manufactured by HP, Pelsee, Azdome, Imagebon and Rexing; Neither of those companies responded to WIRED’s requests for comment.
Although NHTSA acknowledges problems with “some advanced driver assistance systems,” researchers are clear: They’re not sure what this observed hazard light effect has to do with Tesla’s Autopilot problems. “I don’t pretend to know why Teslas crash into emergency vehicles,” Nassi says. “I don’t know if this is still a vulnerability.”
The researchers’ experiments also focused solely on image-based object detection. Many automakers use other sensors, including radar and lidar, to help detect obstacles on the road. A smaller group of technology developers (Tesla among them) maintain that image-based systems augmented with sophisticated artificial intelligence training can enable not only driver-assistance systems, but also fully autonomous vehicles. Last month, Tesla CEO Elon Musk said the automaker’s vision-based system would enable autonomous vehicles next year.
In fact, how a system might react to flashing lights depends on how individual automakers design their automated driving systems. Some may choose to “tune” their technology to react to things that they are not entirely sure are actually obstacles. In extreme cases, that choice could lead to “false positives,” in which a car might slam on the brakes, for example, in response to a cardboard box shaped like a small child. Others may adjust their technology to react only when they are very sure that what they are seeing is an obstacle. At the other extreme, that choice could lead to the car not braking to avoid a collision with another vehicle because it is completely unaware that it is another vehicle.
Researchers from BGU and Fujitsu came up with a software solution to the emergency flashing light problem. Called “Caracetamol,” a combination of “automobile” and the painkiller “Paracetamol,” it is designed to avoid the problem of “seizures” by being specifically trained to identify vehicles with flashing emergency lights. Researchers say it improves the accuracy of object detectors.
Earlence Fernandes, an assistant professor of computer science and engineering at the University of California, San Diego, who was not involved in the research, said it seemed “solid.” “Just as a human being can be temporarily blinded by flashing emergency lights, a camera operating within an advanced driver assistance system can be temporarily blinded,” he says.
For researcher Bryan Reimer, who studies vehicle automation and safety at the MIT AgeLab, the paper points to broader questions about the limitations of AI-based driving systems. Automakers need “robust, repeatable validation” to uncover blind spots like susceptibility to emergency lights, he says. He worries that some automakers are “moving technology faster than they can test it.”