Autonomous Driving and Artificial Intelligence: Continuing

Without artificial intelligence, autonomous driving will not work. Until the man-made brain is so far as to drive the Automobile through the Blechlawine, many hurdles still have to be overcome.

The future scenario of the autonomous driving level five is still many years away. Picture: Bosch
When you look through the press releases of the automobile builders, one comes to the conclusion that it is only a matter of time before the Robo cars roll over the streets without a driver. In 2021, BMW intends to launch an autonomous vehicle on the road with i Next. On request, the münchen clarify: "Level three". This means that the car automatically sets the turn signal, changes the track and overtakes. Especially on motorways. From level five, that is, without steering wheel everywhere, this is still far away. "Let's see if this happens," smiles a managing director of a supplier company, which supplies technology for the guideless cars, to this year number. "The technology is so young that it would be daring today to say when level five of autonomous driving is reached.

The technology with which the self-propelled cars are supposed to do the right thing in every situation is highly complex and still a few years away from the start of production. The step from autonomously driving the level three to the highest level is a huge one. Meanwhile, Audi, Ford & Co. eagerly driving cars that spill thousands of kilometers. Many in California, some in Nevada and BMW soon also in Munich. However, the whole optimistic water level reports of the tests and milestones still leave out a decisive feature: The question of the environment in which these tests take place. In sunny California with good roads and no winter the conditions are ideal. In Cairo or Mumbai, things look quite different - even for level three.

Even driving a highway is a real challenge for the software. GPS is too inaccurate, so different indicators have to be drawn, blue or green signs, structurally separated lanes, the distances of the guide posts, traffic flowing several ways in one direction and Also stored pictures of a highway traffic. The data transfer and the required computing power are immense. "A vehicle will not be able to think completely on its own in the next few years, it needs help from the outside, otherwise the car will be overwhelmed," explains the director of the ZF think tank, Malgorzata Wiklinska. This support could come from an infrastructure that communicates with the vehicle. But that costs money and many years to build up.

Until all cars will communicate with each other (Car2Car) and with the infrastructure (Car2X), years will pass. For the inventors of the artificial intelligence, the car nowadays comes into an "unclean" area, because almost all other vehicles are controlled by people, are so difficult to predict, and above all do not speak with the Robo car. The artificial brain and its senses sensors and cameras must cope with various problems within seconds fractions alone: ​​cross traffic, motorcycles and bicycles running zigzag, pedestrians jumping between cars, and traffic in the dark shadow. Added to this are loading zones for trucks.
Experts go from 25 to 30 years before the cars reach the highest level of self-drive. "It is not possible to get this all the way down to 2021," confirms Elmar Frickenstein. In addition to high-resolution cameras, radar sensors and as many ultrasonic sensors as possible, a computer brain is required to enable cars to drive fully autonomously even in the nastiest tangle of tufts. This is why autonomously operating cars are first used in cities with the best possible traffic flow. Because of the many one-way streets and the relatively evenly flowing traffic are also US-American cities like New York or smaller, such as Pittsburgh.

Other logic

The challenge is: the fault tolerance in the decisions made by the autonomous vehicle must go against zero. After all, it is always about the safety of all traffic users. This makes the task all the more challenging. The way to inspire artificial intelligence to the machine brain is stony. And that is still understated. "We educate the software as a child and say when it has to respond to what situation," says Nvidias Automotive CEO Danny Shapiro. The "Heranwachsende" is fed with millions of pictures and the correct behavior is programmed. These decisions are then stored in huge databases. Time and again, traffic situations are being played through and people are monitoring whether the software responds correctly. The aim is, Which simulates the car perfectly with similar patterns. "These patterns are nothing but human intuition," explains Michael Fausten, who is responsible for autonomous driving at Bosch.
The programming logic of this software differs fundamentally from the familiar "if", "then" as we know it today. This means that the human being pre-sets a solution for the car and stores it in the system. This can not be achieved with autonomous driving and the incredible variety of situations that occur every day in road traffic. "Artificial intelligence has the advantage, due to learned knowledge, to recognize similar traffic situations," explains Dr. Gerhard Gumpoltsberger, Head of Innovation Management at ZF, who works together with Nvidia on autonomous driving. Finally, the skills of the Robo pilot depend on human life. "There is a big difference between using artificial intelligence in a consumer or security-related area.