Artificial Intelligence MINDS IN MACHINES
First fruit of a cooperation between ZF and computing company Nvidia, the ZF ProAI processor also represents a major milestone on the road to automated driving.
For gaming fans, the Nvidia brand has been synonymous with high-performance graphics ever since its products first appeared in 1999. The GeForce 256 was one of the first dedicated 3D graphics processors (GPU) for computer games. Since then, Nvidia has succeeded in doing what no other company apart from Intel has done: turn an innocuous piece of hardware tucked away in the innards of the system into a high-value brand item. Today’s PC desktop builders use the latest generation of GeForce processors as a major selling point for their systems.
Meanwhile GPUs have powered their way into new fields, far removed from computer gaming. Nvidia CEO Jen-Hsun Huang believes his technology will quite simply “reinvent our world.” That’s because the GPUs that Nvidia has developed for processor-intensive 3D gaming over the last 20 years are now playing a key role in what is probably the most important IT development since the invention of the internet: artificial intelligence, or AI.
Video: Opportunities through Artifical Intelligence
Deep Learning: full speed ahead!
AI is already used in many applications, such as smartphone voice control software capable of adapting to user input. The very latest innovation is “Deep Learning”, a way of enabling AI systems based on manmade “neural networks” to learn without any human intervention at all – something for which conventional computer architectures simply aren’t fast enough. But the calculations needed to train an AI system are not dissimilar to those required to generate 3D worlds in computer games, based on mathematical models that must be reiterated billions of times. And that’s where Nvidia comes in.
“Neural nets fitted with our GPUs can be trained orders of magnitude faster,” explains Jaap Zuiderveld, head of Nvidia’s European operations. This discovery is laying the foundations for an application that will transform all our lives: automated driving. In order to make correct decisions in heavy traffic, self-driving vehicles must process vast quantities of data in real time – data streaming from dozens of cameras, as well as lidar, ultrasound and radar sensors. At the same time, self-driving vehicles are constantly learning, with every mile teaching a new lesson. And finally, they’re sharing their new experiences with each other via the Cloud.
Now ZF is working with Nvidia to develop a system that will make artificial intelligence available to the mobility industry. At the Consumer Electronics Show (CES) 2017, ZF unveiled the company’s first system for automated freeway driving, based on Nvidia’s AI. ZF ProAI enables vehicles to “understand” their surroundings by using deep learning technology to process and interpret data from sensors and cameras. The two partners are also working on solutions for highly automated and fully automated driving.
At the heart of ZF ProAI is a processor developed specifically for automotive applications – simply installing an ultrafast PC processor in a car isn’t an option, as Torsten Gollewski, head of ZF Advanced Development, explains. “Temperatures in automotive environments range from minus 30 to plus 80 degrees Celsius. Other stress factors include humidity, vibration and high G-forces – conditions which conventional PCs and games consoles aren’t designed to handle.”
ZF supplies ZF ProAI as a system that can built into a vehicle, updated via the Cloud, and upgraded with additional functions throughout the lifetime of the vehicle. Series production of the new ZF ProAI system is scheduled for 2018.
But the benefits aren’t exclusively reserved for cars. The key concept here is “automated operations”, based on the realization that an electronic component capable of handling the stresses and strains found in passenger cars is also ideal for use in products in other industries. “Working with Nvidia, we’re bringing the supercomputer standards of performance required for artificial intelligence not just to cars and commercial vehicles, but also to all kinds of industrial applications,” confirms Dr. Stefan Sommer, ZF CEO.
Just what Nvidia’s technology is really capable of was demonstrated by company founder Huang in April of last year in San José, California. He fed 20,000 images of paintings from the Romantic era into an AI-based computer. Once this had been done, the Deep Learning machine proved capable of creating an original picture on its own, albeit a picture that was clearly in the same style as those it had just “seen”. Although the source images were categorized, the computer was not given any information on the subject of each painting. It had to work that out for itself.
Huang believes this development is the start of a new age. “We believe that Deep Learning represents a whole new computing model. The results are quite simply superhuman.” ZF developer Gollewski views the two companies’ collaboration as an opportunity to make a quantum leap forward: “Our cooperation with Nvidia gives us access to a world of digital possibilities in a totally new dimension!”