Found another way to cheat Tesla autopilot

Found another way to cheat Tesla autopilot

October 15, 2020 0 By autotimesnews

Machine vision can be confused with short-term images

Scientists at the Ben Gurion University in Israel, who have been working on identifying vulnerabilities in machine vision technologies for the last two years, have published the results of their research. In their opinion, the Tesla autopilot can be confused with the help of electronic road signs, if you run the necessary information on them.

Tesla’s autopilot works in a similar way to human vision. An electric car is guided by what the autonomous system sees around it. However, sometimes the focus of machine vision is information that can confuse the electric car and force it to make an unexpected decision, which in turn can provoke a dangerous situation on the road.

Confusing factors can be electronic billboards that display advertisements or warning messages. Tesla electronics need only 0.42 seconds to notice a sudden image. If at the moment of movement of the electric car on the boards the speed limit signs or stop orders light up for a moment, the autopilot can react to them with sudden braking, even if the electric car is driving at high speed. In addition, if a picture of another car or a pedestrian is displayed on electronic boards, the Tesla autonomous system can take them for real conditions and try to avoid an accident.

According to scientists, such a vulnerability can be exploited by hackers. After all, road electronic boards do not have a high level of protection and can be used by intruders to create emergency situations on the road. In addition, Israeli experts emphasized that the autopilot system should be replenished with a lidar so that the electronics can analyze a larger number of external factors and not get lost when flat images appear that the optical range finder can recognize.

Israeli experts have also suggested using their own algorithm called “Ghost Hunter” in autopilot. The system collects all data provided by radars, lidars and sensors and automatically recognizes short-term flashes, filtering them out of the real traffic situation.

At the end of September, a video was posted on the Web, filmed by the on-board cameras of Tesla Model 3. The presented footage recorded an incident in which the autopilot system of an electric car managed to avoid a collision with a truck.