Intelligence tests: The new battle ground for driverless car experts?
NVIDIA wants to build the smartest computer for automated cars: The central driver assistance control unit (zFAS) in the new Audi A8 is only the beginning.
In the latest Spider-Man movie, viewers get a sneak peek of what the new Audi A8 can do. And while it might not pack the punch of his superhero colleague’s new Batmobile, there are certainly a number of features that catch the audience’s eye. Take the button marked AI Traffic Jam Pilot for example. Young hero Peter Parker presses the button during his driving test, before taking his hands off the steering wheel. "I guess turning the wheel is too much for you kids nowadays, huh?" mutters the unnerved instructor. As Parker goes on to park the car automatically – foiling a couple of bank robbers in the process – one thing is clear: superheroes are about to get new wheels, and artificial intelligence will be a central element.
Until recently, humans were essential for driving assistance systems – both as operators and as backups. Now, even luxury limousines can function without drivers for significant amounts of time. Level 3 automation is being premiered in the new A8. The fail-safe system has to create an environmental model, plan a route and choose a driving strategy – almost in real-time. This is why the most important function, the AI Traffic Jam Pilot, refers to artificial intelligence. It is meant to be a self-learning system that does not immediately shut down when unexpected situations occur. Like a learned student, the clever computer should be able to understand constellations that were not previously programmed. At present, the robocar is operating within clearly defined limits: the system is restricted to driving at a maximum speed of 37 mph (60 km/h) and automatic parking.
The Traffic Jam Pilot - a little bit of everything
The signals that the Traffic Jam Pilot needs for highly automated driving come (among other things) from Audi’s central driver assistance control unit (zFAS). The high-tech central controller integrates high performance processors from NVIDIA (Tegra K1), Altera (Cyclon V), Infineon (Aurix) and the EyeQ3 image processing system from Mobileye. Whenever the Traffic Jam Pilot is active, the driver can remove his hands from the steering wheel and devote himself to other activities. This is what makes the new A8 a significant step on the road towards autonomous driving. The prerequisite for this is a complete and continuous assessment of a car’s surroundings. NVIDIA talks of data being processed at a rate of 2.5 billion inputs per second to create a consistent environmental model. Input is provided by an extensive set of sensors: twelve ultrasonic sensors, six cameras, five radars and (for the first time) a front-mounted laser scanner.
Who else is swotting up on AI?
The tablet-sized zFAS is undoubtedly a milestone; but others are already gearing up to overtake. During the Bosch Connected World conference in March 2017, NVIDIA CEO Jen-Hsun Huang announced a new generation of in-car supercomputers. As stated in a press release, NVIDIA highlighted that ADAS systems are a long way from a self-driving car. The amount of processing required for an autonomous vehicle is orders of magnitude greater. Huang noted the incremental amount of processing to be at least 50 times greater. This is why NVIDIA – together with Bosch – is planning to develop a supercomputer for the mass market. The collaboration will use NVIDIA DRIVE PX (an open AI car computing platform), along with NVIDIA’s forthcoming Xavier technology. It will in fact be the first time the technology has been incorporated in the platform: "Xavier can process up to 30 trillion deep learning operations per second and only requires 30 watts of energy," Huang explained at the event. Using the new technology, NVIDIA wants to clear the way for Level 4 automation by the end of 2018.
One of the first companies to benefit from the DRIVE PX platform will be Mercedes in a joint project with Bosch. The Stuttgart-based technology firm is using the slogan Das Auto kommt zum Fahrer (the car comes to the driver) to communicate the "partnership at eye-level". As Stephan Hönle, Senior Vice President of the Business Unit Automated Driving at Bosch explains: "This is not a classic tier-1 company relationship, but rather a development cooperation between the world's leading supplier and the world’s leading premium brand.” Alongside their collective technological expertise, both partners are also able to invest several hundred million Euros in development costs. The goal: driverless robo-taxis, with Level 5 automation and no steering wheel. They will operate in urban areas, reaching speeds of 47 mph (75 km/h). According to Bosch, the first pilot projects will start at the end of next year.
Succeeding in the robocar race: A matter of experience?
So it seems things are looking good for full automation – and for NVIDIA. Since 2011, the relatively young manufacturer of graphics processing units has developed into a rising star of the car industry. Diligent Huang continues to build new automated-driving partnerships: first Tesla and Toyota, more recently Volvo and Volkswagen. NVIDIA wants their technology to become a universal platform for automated driving.
During the International Conference Advances in Automotive Electronics (Elektronikkonferenz) in Ludwigsburg, Germany, at the end of June, Huang presented some clear figures: In a robo-taxi operating at Level 5 with 25 sensors, a central computer would require 1000 watts of energy if graphics processors were being used in parallel. In the case of conventional CPU architecture, the NVIDIA boss estimated that 3000 watts would be needed to generate the same power. Therefore, in the latter case, central computers in electric cars would become huge energy guzzlers, significantly reducing their reach. Through this statement, Huang inadvertently referenced discussions about processors in mobile phones. Intel, the former market leader for high-performance CPUs in PCs, has had problems keeping up with the new market boom for mobile internet. Is this now set to happen again with the ultimate mobile device: electric cars?
The statements made in Ludwigsburg did not impress Head of Electronics Development at BMW, Christoph Grote. In close cooperation with Intel, BMW wants to bring the BMW iNext to the market by 2021. Grote hinted that the Intel platform would use parallel processing semiconductors for automated driving. Following the takeover of Altera two years ago, Intel also has the necessary knowledge. However, in the end, it is not only the computer architecture that will be decisive, but also refinement in its application. In fact, auto experts often find that NVIDIA programmers can make maximum use of processing power when using parallel computing. Whereas, due to their current lack of experience (vis-a-vis their counterparts from NVIDIA), software developers from the auto industry only reach 50 percent utilization rate.
Without a doubt, semiconductors and the software architecture based on them will play a decisive part in the outcome of the robocar race.
Will the future of driverless mobility depend on semiconductors and software? What do you think? Let us know in the comments!