Autonomous cars are known for being slow, but certain situations on the road require the ability to quickly react and perform intricate maneuvers. How can self-driving cars pass these tests? Scientists from Stanford University decided to find an answer for this question.
The solution they proposed was a neural network that would enable the autopilot to perform high-speed maneuvers rivaling Formula-1 race drivers.
The program was trained using 200 thousand examples of driving in different weather conditions, on various types of terrain. Afterwards, the autopilot program was tested on the Thunderhill race track in Sacramento.
The cars used in the experiment included Volkswagen GTI with a Niki autopilot and Audi TTS equipped with a Shelley autonomous driving system. Both vehicles demonstrated impressive results that were comparable to a real race car driver’s work. Although information about the Thunderhill race track was uploaded to the Shelley system, Niki had to make decisions without any information about the road.
Share this with your friends!