The auto industry seems to announce things in waves, and after a lull, a torrent has sprung forth. The largest breaker comes from Nvidia, which unveiled the next generation of its Drive PX sensor fusion computers – codenamed Pegasus. However, Intel hopes to make a splash with a new mathematical formula to help self-driving cars never crash, and ARM is racking up wins for its new Cortex-R52 SoCs, with Denso announcing that it had taken out an extensive license.
Nvidia seems to have significantly shrunk the size of the computer, with the on-stage presentation at its GTC event in Munich claiming that Pegasus is roughly the size of a license plate – and capable of providing the compute power for SAE Level 5 operation. That’s enough for no human to be present in the car, and for the car to have no operational equipment (wheel, pedals, dials) inside – but the claim of Pegasus having “the AI performance of a 100-server data center” sounds a little market-y.
The Pegasus system will house two of Nvidia’s Xavier SoCs, flanked by two ‘next-generation’ GPUs, based on the new Volta design, and will apparently be capable of ten-times the compute power of the Drive PX2 predecessor – clocking in at 320 trillion operations per second (TOPs).
By comparison, the Intel/Mobileye EyeQ5, currently in development, claims around 12 TOPs, and so it seems that Nvidia is going to trounce the EyeQ5 in a synthetic benchmarking test – but these numbers do not translate nicely into real-world performance, and so we have to wait and see how cars built on top of Pegasus actually perform compared to the Intel/Mobileye equivalent.
As it stands, Nvidia claims to cover all bases in self-driving silicon demand – with single mobile chips ready to power SAE Levels 2 and 3, and the new Pegasus design ready for Level 5. It further argues that its open software architecture and offerings mean that automakers and Tier One providers can reduce their development times, by using things like AutoCruise and AutoChaffeur, which are part of the DriveWorks suite.
Notably, the announcement talks less about cars in general, and is instead quite specific about ‘robotaxis,’ – arguing that they will save millions of hours of lost driver time. While we’re fairly sure Pegasus could power a regular privately-owned car too, Nvidia is keen to note that the computing demands of driverless vehicles are still around 50-100x that of the most advanced cars on the road today.
Nvidia also announced that DHL and ZF (a Tier One automotive supplier) planned to deploy a test fleet of autonomous mail trucks in 2018, using ZF’s ProAI system (based on Nvidia’s Drive PX boards) for carrying out the last-mile deliveries. An existing fleet of StreetScooter trucks, of which DHL currently has 3,400, will be upgraded with the boards, cameras, LiDAR, and radar, and will be aided by a DHL data center using Nvidia’s DGX-1 AI computer to train its neural networks.
Moving over to Nvidia’s chief rival in the sector, Intel (which spent some $15.3bn to acquire a powerful position in the market) is hoping that its new Mobileye assets build on the success of previous Mobileye designs (the EyeQ2 and EyeQ3) and continue to enjoy wins with automakers and their suppliers.
This week, Intel and Mobileye have published an academic paper that contains a formula that Intel says provides a mathematical model that will “ensure a self-driving vehicle operates in a responsible manner, and does not cause accidents for which it can be blamed.” You can find the full paper here, and a layman’s version here.
The main thrust of the publication is the proposed Responsibility Sensitive Safety model, which aims to create a mathematical version of ‘responsibility and caution’ and define a ‘safe state’ where a self-driving car can’t be the cause of an accident, regardless of what other vehicles around it are doing. Basically, it’s looking to ensure that the manufacturers are never going to be the target of a humongous lawsuit, and these are the mathematical rules to follow to adhere to Intel’s guidelines.
It does make for quite an interesting read, and as it doesn’t rely on specific technologies or intellectual property (currently, at least, we believe), it could be adopted widely by the auto industry. Lawmakers and insurance companies will also be rather interested in the approach, as those two topics can spark nearly endless debate about challenges ahead for integrating autonomous vehicles on today’s roads – and if the model works as functioned, then it could iron out rather a lot of the concerns that are frequently expressed.
Elsewhere in the industry, Qualcomm’s acquisition target NXP is preparing to unveil a ‘whole vehicle’ platform to simply its myriad of automotive processors. Called NXP S32, it is apparently being designed to that the OEMs and their Tier Ones can better re-use code and software across the vehicle domains – such as drivetrain, braking, stability management, HVAC, etc.
We will have to wait to see the fruits of those labors, as NXP doesn’t plan on releasing anything on that new platform until 2018. Qualcomm should have completed the acquisition by then (or the financial analysts are going to have a field day), but NXP will be wary that the likes of Intel and Nvidia are aiming to muscle in on a core market – as NXP is still the leading automotive chip supplier. Speaking to EE Times, NXP said that 8 of the top 15 OEMs are already adopting S32.
And the final piece of notable automotive news was that Denso, another major supplier, has taken a license for ARM’s new Cortex-R52 chips, which it plans to use to create processors that comply with the very strict safety regulations. ARM says that the R52 is being rapidly adopted by “the worldwide design community as the go-to solution for embedded functional safety.”
“Partnering with Arm gives us access to its vast ecosystem and will allow us to continue to drive toward enabling the next generation of safe, highly automated driving systems and vehicle control,” said Hideki Sugimoto, Project GM, BASIS Electronics R&D Div. of Denso. “As the automotive revolution continues, our focus is on working with partners to build the right technologies to meet changing industry needs. We evaluated the Cortex-R52 and concluded that the processor’s functional safety capabilities were an excellent match for advanced future automotive applications. Denso, along with our newly established subsidiary NSITEXE, will now work to further enhance these solutions.”