You see artificial intelligence (AI) in action every day when Google displays products based on your search behavior or YouTube recommends interesting videos to you. The importance of this theme of the future for automakers and its priority in this high technology area are less obvious. The vast number and complexity of requirements related to automated driving will place the Volkswagen Group on the cutting edge of research into artificial intelligence.
“Automated driving is pretty close to the things that people used to imagine when they thought of the term ‘intelligent robot.’ A car that is driving itself down the street faces an enormous number of challenges created by the speed of processes and the variety of situations it faces.”
The complexity involved in mastering traffic exponentially exceeds the requirements placed on autopilot systems in aircraft or driver-less subway trains. Another major problem is the fact that not all drivers always follow the rules of the road. For this reason, automated cars need a software that can completely and reliably react to unexpected situations: This means one thing: The car must learn to think on its own.
“Our work is based on collecting as many experiences from real driving situations as possible. This information also forms the foundation of artificial intelligence processes that we are implementing with increasing frequency. Our goal is for artificial neural networks to gain knowledge about the real world in development operations.“
One of the many research projects being conducted into automated driving involves simulating traffic conditions in major metropolitan areas of China. Drivers faces particularly complex traffic situations there. These situations can be translated into a virtual reality in which computers create additional scenarios.
Surprisingly, video games are proving to be a big help in terms of hardware. Graphics processes play a key role in the development of automated driving because they can already quickly and simultaneously perform a number of different tasks at a relatively low price.
Helge Neuner: “One major issue is estimating the size and distance of objects, as well as identifying them. We can learn a lot from people here.” As a result, another research project is working on the coordination and redundancy of various measurement systems. To gain an exact picture of their environment, computers use several different types of measurements that can be used to offset each separate system’s own particular weaknesses. Cameras can identify objects well, but are unable to exactly measure the distance to them. Radar, on the other hand, can measure precisely, but displays shapes inadequately. Once the different systems are able to flawlessly communicate with one another, they will be far superior to humans. The reason is simple: A 360-degree view of a vehicle's surroundings will be created, and the sensors can also “see” at night and in fog.
Another spectacular research project is called Race Pilot. In this work, engineers are pushing automated vehicles to their limits on closed tracks.
“At first glance, Race Pilot seems very extreme. Of course, an automated vehicle does not have to travel faster than a race car. What we are trying to do is to safeguard the processes when they reach these upper limits and enable the vehicle to outperform the average driver in critical situations.”
To turn the car into a thinking and driving object, far-reaching improvements must be made in today’s electronic systems. Nonetheless, researchers believe that far-reaching extra effort will ultimately produce benefits for the customer. First, Moore’s law – the continuous improvement of computer components’ performance at ever lower costs – will help clear the way. For the equally expensive sensors, mass production will lead to a reasonable price level, just as it has previously done for many other safety innovations in cars. At the same time, automated driving will make new, attractive forms of individual mobility possible.