Last week, we reported that many advanced driver assistance systems (ADAS) had trouble recognizing pedestrians wearing reflective clothing after dark. A new study shows that they may also have trouble with flashing lights from emergency vehicles, effectively having a “seizure,” according to the study’s authors.
Japanese tech company Fujitsu and Ben-Gurion University of the Negev in Israel conducted the study, which identified 16 instances of Tesla vehicles crashing into emergency vehicles while using the automaker’s Autopilot feature. Researchers tested ADAS cameras from several manufacturers and found that the systems displayed significantly lower object detection confidence when lights flashed nearby.
Calling the problems “digital epileptic seizures,” the study coined the phrase “EpileptiCar,” pointing out links between the systems’ behavior and epilepsy in humans. Like with the human eye, flashing lights change how cameras “see” objects, and the study found that cameras from all five major manufacturers suffered similar issues.
The study noted that more research and development is needed to push them closer to a human’s level of object recognition. Artificial intelligence can help but lacks the necessary reasoning skills to cope with complex situations.
These results cast doubt on autonomous vehicles that rely primarily on cameras to map the world around them, including Tesla’s Full Self-Driving feature, which is based on camera input. Additionally, the study’s authors said that bad actors could take advantage of the cameras’ weaknesses, flashing lights to cause crashes.
[Images: Delbo Andrea, Scharfsinn via Shutterstock]
Become a TTAC insider. Get the latest news, features, TTAC takes, and everything else that gets to the truth about cars first by subscribing to our newsletter.
