Self-driving cars have been driving on our roads already for a long time. They use video cameras and special sensors called lidars to see objects around them, computer vision to understand what those objects mean and an AI autopilot program that controls the driving. To train such a self-driving AI program, the car first drives with a human driver for many years and learns from them. After that, it can drive itself quite safely - you simply tell it where you want to go and you sit in the back seat. Statistically, AI autopilots are involved in accidents significantly less often than humans, but why are self-driving cars still so rare? Because AI autopilots, although they make the right decision 95% of the time, can make a mistake 5% of the time, even in very simple situations in which a normal driver would never make a mistake.
Add comment
