Artificial intelligence is the auto industry’s big trend. Adaptive systems – capable of what software developers call “deep learning” – will make future traffic safer. And Continental is among the companies spearheading the development process.
A black Pontiac Firebird Trans Am races through a small American town, threading through narrow streets at top speed, blasting past stunned pedestrians before exiting onto the highway. But there’s no driver behind the wheel – the black sportscar is driving itself, scanning its surroundings to identify potential obstacles or hazards and reacting accordingly. The car turns off the highway at the next exit, swerves into a residential area and comes to a sudden stop. Its owner comes running from across the street, jumps into the driver’s seat and grabs the wheel, and together they disappear around the corner of the next block… It’s a scene from cult TV series “Knight Rider”, filmed back in the 1980s, in which protagonist Michael Knight fights evil with the help of K.I.T.T., his talking, self-driving car.
Back then, such a scenario was pure science fiction – now, it’s very close to becoming reality. For years, carmakers have been testing autonomous vehicles capable of driving through city centers, down country roads or along freeways with no human intervention. A self-driving car must be capable of responding immediately to whatever happens – for example, by initiating emergency braking if a pedestrian steps into the road, or finding an alternative route if a bridge is closed. Circumstances and situations that may occur along a route from A to B are as complex as they are unpredictable. It’s quite literally impossible to pre-program every single step in a vehicle’s responses to every conceivable traffic scenario. Which leaves only one solution: deep learning.
First of all, the car’s control system stores examples of human driving behavior and then attempts to imitate it – in other words, it is familiarized with human norms. At this point, it’s no longer simply obeying programmed commands – it’s learning on its own, just like a human. The car goes to driving school, learning more every time it takes a driving lesson. This is made possible by self-learning computer chips. “There’s no way in the world you can program everything required to manage a vehicle in everyday traffic,” says Scott Keogh, currently still in charge of Audi’s U.S. operations, but due to move over to head Volkswagen USA on November 1. “The only way out is to use artificial intelligence.”
Artificial intelligence (AI) is today’s big trend. Cars don’t just drive – they also think. All the major carmakers and leading auto suppliers are currently researching systems that will make vehicles smart. Audi and Mercedes, for example, are developing prototype vehicles that will be able to drive themselves using self-learning computers. Meanwhile Volkswagen is concentrating on the interior, developing a cockpit that works with the help of artificial intelligence. All these efforts are focused on nothing less than transferring the human brain’s neural network into a computer-controlled system. “Neural networks are one of the keys to mastering the challenges of autonomous driving,” explains Dr. Uwe Franke, who heads up Daimler’s Image Understanding group. In other words, “What we’re seeing is the growing interpenetration of biology and technology,” says Professor Reimund Neugebauer, President of the Fraunhofer Society. “You could call it the ‘biologization’ of industry.”
The key role in this “biologization” is played by chip manufacturers, who work closely with automakers and their suppliers. Semiconductor manufacturer Nvidia is developing chips on which to store the vehicle-movement algorithms generated by machine-learning processes. Nvidia has already demonstrated how this works in an Audi Q8. Millions of images of obstacles were loaded into the AI system installed in the car. Using a combination of driving simulations and real-world driving maneuvers, the system was then trained to brake every time it encountered these types of obstacles. The Nvidia system remembers the connection between identifying a particular obstacle and the subsequent braking maneuver, and eventually becomes capable of stopping on its own in new – but similar – situations, without requiring a man-made algorithm for every single permutation.
The enormous potential of artificial intelligence is also reflected in figures published in a study by the Fraunhofer Society. By 2025, the prospective market for semi-autonomous vehicles alone should be worth around 36 billion dollars; for self-driving cars, it might be worth as much as eight billion dollars. In 2014, the premium segment comprised some 0.6 million fully or partially automated vehicles; by 2035, this figure is expected to rise to at least 10 million units. Over the same period, the number of fully or partially automated high-volume vehicles should rise to 38 million. “Cognitive machines will play a vital role in transforming business. These technical systems are self-learning and capable of applying what they learn to new situations. They can plan processes, make predictions and even interact with people,” says Fraunhofer boss Neugebauer.
Continental is a leading developer of artificially intelligent systems. To bring the digital world’s user experience (UX) into the car, Continental is using artificial intelligence-based concepts to transform the entire vehicle into a digital companion. Using deep learning algorithms to observe and interpret user behavior, this digital companion is able to adjust navigation and infotainment options accordingly, and even anticipate drivers’ wishes. Continental is now extending its international artificial intelligence network to include Silicon Valley. “We’re joining forces with the world’s leading AI researchers,” says Demetrio Aiello, who heads Continental’s globe-spanning AI Advanced Development group in Regensburg. “To supplement our partnerships with Oxford University, the German Research Center for Artificial Intelligence and other institutions, we’ve now signed a five-year contract with the University of California’s Berkeley DeepDrive (BDD) AI research group.” Among other things, the joint venture will focus on optimizing the speed of in-vehicle neural networks and securing AI systems in safety-critical applications. The main objective of the partnership is to feed the findings of this AI research into series production as quickly as possible. During the partnership’s first year, Continental will prioritize two of Berkeley DeepDrive’s areas of research in particular. First, the “testability of AI algorithms in safety-relevant systems”, because drivers must be able to relax in the knowledge that their vehicle’s complex technology works reliably in everyday operation. With this in mind, BDD is developing better methods for testing AI systems’ dependability. And second, the Berkeley researchers are investigating ways of making AI applications exceptionally memory-efficient, with the aim of speeding up and optimizing neural networks. This will make it easier to use AI-based methods in vehicles at a later stage.
To coordinate all these AI-related research activities, back in 2015 Continental launched a centrally managed AI Advanced Development group. The technology company is working with Nvidia, Baidu and many other research institutions specializing in the field, such as Oxford University, the Technical University of Darmstadt, the German Research Center for Artificial Intelligence (DFKI) and the Indian Institute of Technology Madras (India). In May 2018, Continental’s Driver Assistance Systems business unit opened a Deep Machine Learning center of excellence in Budapest (Hungary). By the end of 2018, Continental will employ some 400 engineers specializing in AI around the world.
Clearly the number of computer chips in vehicles will rise dramatically in the future. At present, the average value of the computer chips in a vehicle comes to around 330 dollars. “In a self-driving car, we’d expect to add another 500-700 dollars to that figure – maybe even as much as 1,000 dollars per vehicle in specific cases,” says Reinhard Ploss, CEO of semiconductor manufacturer Infineon. All this activity is delighting software developers. Ten years ago, companies such as Nvidia and Infineon mainly produced chips for computer products outside the automotive industry; since then, the share of added value generated by the auto industry has risen sharply. “Half of the growth forecast for our automotive division will come from e-mobility, driver assistance systems and autonomous driving,” says Ploss.
But semiconductor manufacturers aren’t the only ones to benefit from this trend; the carmakers themselves are feeling the love. The important thing, however, is to be the first – or one of the first – to bring new systems to market. Because a new study by management consulting firm McKinsey suggests that two thirds of car drivers in Germany, the U.S. and China would willingly change brands to take advantage of more sophisticated artificial intelligence features. And so Audi, Mercedes and the rest are not skimping on investments; the same McKinsey study indicates that the industry has invested more than 51 billion dollars in AI over the past seven years.
So the fictitious Knight Rider scenario is closer than ever. “The next three or four years will be pivotal,” asserts Johann Jungwirth, Chief Digital Officer at Volkswagen – he even talks about “reinventing the car”. In the cockpit, for example, “you often have to press seven or eight buttons on your in-car entertainment system before you reach the menu item you want,” he continues. “We want to slim that down to one – or ideally, none at all.” The car of the future should be able to recognize a driver’s facial expressions, mood and intention, he adds, and by combining all this with positioning data and driving behavior, select the best route to match. “In the old days, the engine was the heart of a car,” says Jungwirth. “In the near future, the autonomous driving system will play that part.”