Autonomous Driving

Autonomous driving means: a car drives itself, without a human steering or braking. The vehicle perceives its surroundings via cameras and other sensing devices. A computer in the car evaluates this data and decides what should happen next. It thus recognizes pedestrians, traffic lights, lane markings, and other cars and reacts to them. Experts distinguish between six levels, from 0 to 5. At level 0, the human does everything themselves; at level 5, the car theoretically wouldn’t even need a steering wheel anymore.

What’s at stake: safety, money, and entire professions

Worldwide, more than a million people die in traffic every year. By far the largest share of these accidents is caused by human error: distraction, fatigue, excessive speed. A machine doesn’t get tired and doesn’t look at its phone. This is precisely the basis for the technology’s great promise. Whether autonomous cars are actually safer in practice, however, has not yet been conclusively proven.

Economically, enormous sums are at stake. In a taxi, the driver’s wage makes up a large part of the fare. If it disappears, rides could become significantly cheaper. Companies like Waymo, Tesla, or the Chinese company Baidu have therefore been investing billions for years. For investors, autonomous driving is one of the reasons why individual automakers are, at times, valued on the stock market like technology companies.

At the same time, tough legal questions arise. If a self-driving car causes an accident, who is liable? The owner, the manufacturer, or the programmer? In Germany, a law has existed since 2021 that permits driverless vehicles under strict conditions. Nevertheless, the debate about it is far from over.

From sensor to steering motion

The first step is perception. Cameras provide images, radar measures distances and speeds via radio waves. Many vehicles additionally use lidar, a method using laser pulses that maps the surroundings three-dimensionally. From all these sources, the onboard computer builds a continuously updated model of the environment. A neural network — that is, a computer program trained from examples — assigns meanings to the recognized objects: child, cyclist, plastic bag.

The second step is prediction. The system estimates where each recognized object will move in the next few seconds. Only after that comes planning: which lane, which speed, which distance? In the end, electronic control units translate this decision into steering movements and brake pressure. The entire cycle runs several times per second.

The difficulty lies not in the normal case, but in the exceptions. Experts call them edge cases: a construction site with contradictory markings, a police officer directing traffic by hand, heavy snowfall on the camera. A human improvises in such situations. A system that only knows learned patterns can fail at this. That’s why test vehicles rack up millions of kilometers before a function is approved for release.

Robotaxis, driver assistance systems, and a look at the news

In everyday life, you usually encounter the technology in small doses. Adaptive cruise control, lane-keeping assist, and automatic emergency braking belong to level 1 and 2. The human remains fully responsible and must keep their hands on the steering wheel. The term autopilot, used by some manufacturers, is therefore misleading.

True driverless cars currently only operate in defined areas. In San Francisco, Phoenix, and several Chinese cities, you can order robotaxis via app. On German roads, this is still a thing of the future, though certain models are allowed to drive themselves temporarily in highway traffic jams.

The term regularly appears in financial news when licenses are granted or revoked again after accidents. Such reports noticeably move stock prices. A common misconception here: a system that works perfectly on a sunny suburban street is far from ready for a European winter.