Machine learning technology is changing every facet of our lives. It will soon determine whether or not we need to drive ourselves.
A car without steering wheel and pedals. Instead, robots should take over the wheel. That is the vision of the car manufacturers. They are using machine learning to disrupt the future of the automotive industry.
“The vehicle of the future is an autonomously driving computer on wheels,” says Elmar Degenhart, CEO of the automotive supplier Continental, who is working on such vehicles together with the US chip manufacturer Nvidia. Connected with the vision is the hope of bringing more safety to the road. 80 percent of all accidents today are caused by human error.
Audi has already brought out a vehicle that drives Level 3 autonomously. Big data and machine learning advances have made this truly possible. This means that the driver is allowed to take his hands off the handlebars on easy stretches of road, for example on the highway. Last year, the German government also created a legal framework for this by making it possible, to put it simply, for machines as well as people to be allowed to drive the vehicle – at least temporarily on the autobahn and on simple stretches of road. In addition to Audi, German manufacturers Daimler and BMW also want to introduce Level 3 vehicles by 2021 at the latest.
The “mastermind” in the car – with intuition from machine learning advances
However, years will pass before the cars can drive completely autonomously and make decisions on their own. Machine learning has not evolved at the pace needed to make that possible, yet.
So far, no automaker has announced that it has developed a production-ready driverless car. It will take some time before fully autonomous Level 4 or driverless Level 5 is reached. “Nobody can currently tell you with certainty when vehicles will be fully autonomous,” Christian Müller told NGIN Mobility and the start-up scene. He heads the Competence Center for Autonomous Driving at the German Research Center for Artificial Intelligence in Saarbrücken.
But the foundation stones are already being laid today. A prerequisite for the Level 5 vehicles is that the vehicles can make decisions completely on their own. This requires the use of new technologies. Joachim Langenwalter is certain that the security level and performance of self-propelled cars cannot be achieved with conventional approaches to object recognition. He is Director of Automotive Software at chip manufacturer Nvidia. The company is working with 370 companies worldwide to develop the technology for robotics, trucks and delivery vehicles on its “Drive” platform. The most prominent partners are Daimler and Bosch.
The manufacturers’ vision is that in the future, cars will be able to learn on their own, comparable to a human being. The researchers summarize the methods behind this under the heading of artificial intelligence. This involves creating so-called neural networks that resemble the human brain. They consist of simple, trainable mathematical units and establish connections between the car’s environment and the on-board computer. These procedures are called machine learning or deep learning.
“For example, the properties of an image showing a stop sign are broken down and “examined” by the neurons,” explains Nvidia manager Langenwalther. “This concerns shape, color, size and so on.” The task of the neural network is then to determine whether or not the image is a stop sign. Based on the incoming information, the system evaluates how likely it is that it is a stop sign. “Deep learning consists of two phases: training and inferencing.” For the stop sign example, this means The more images of it are used to train the network, the better it will be at identifying them correctly. So equipping cars with AI is a bit like teaching students.
Complaints about the autonomous Waymo fleet
Thus, the new procedures are fundamentally different from the if-then principles programmed today: Instead of a programmer finally determining the interpretation of the data, the system makes its own decisions, comparable to human intuition.