Nvidia Deals Tilt Robo-Car Race
6/27/2017 11:51 AM EDT
PARIS — Claiming that robo-car development by automakers has already moved from the R&D phase to production, Nvidia this week unveiled three new partnership deals — all aimed at leveraging its AI car-computing platform.
Nvidia announced Monday (June 26) that Volvo and Autoliv have selected Nvidia’s Drive PX 2 for production of self-driving cars in 2021.
Nvidia said that it also sealed a deal with ZF & Hella, who are both committed to working with Nvidia to deliver with the New Car Assessment Program (NCAP) safety certification for the mass deployment of self-driving vehicles.
But wait. There’s more. Nvidia also disclosed an agreement Volkswagen, under which the German carmaker will expand deep learning “competence” throughout the enterprise and developing a number of AI apps running in the data center.
Prior to these disclosures, Nvidia had already picked up a number of other notable car OEMs and tier ones as partners for autonomous vehicle development. Among them, Tesla has been already using Drvie PX in its current-generation cars. Audi is planning to deliver Level 4 cars based on Drive PX platform in 2020, and Toyota will use Nvidia’s platform to power advanced autonomous driving systems. Separately, Daimler, Mercedes Benz (owned by Daimler) and tier one Robert Bosch have also chosen Nvidia as their autonomous platform partner.
See the relationship map below.
(Source: EE Times)
Nvidia's senior automotive director Danny Shapiro told reporters, “The momentum of autonomous vehicles” is growing. The focus of activity is shifting from development to the “production phase.”
Noting 225 different “engagements” involving its Drive PX platform, Shapiro said that Nvidia is working with “a whole spectrum of players in the automotive industry ranging from OEMs, tier ones to trucks, HD mapping companies, sensors and startups.”
Citing Nvidia’s deal with Volkswagen, Shapiro said AI is now being applied not just to vehicles (i.e. path planning), but also to a backend system where a data center crunches out traffic patterns, flows and driving behavior while looking for anomalies — figuring out the entire transportation ecosystem.
Luca De Ambroggi, principal analyst for automotive electronics at IHS Markit, agreed. “This is the power of Nvidia offering a well distributed and consistent ‘solution’ in different domains from the ‘edge’ up to the infrastructure," he said. "A lot of money (for the OEMs) stays in functionalities like predictive diagnostic, and maintenance, cybersecurity and traffic management.”
Previously some skeptics noted that Nvidia’s AI platform might be effective for research, but not necessarily for production cars. However, Nvidia appears to be defying those predictions.
Ian Riches, director of global automotive practice at Strategy Analytics, told EE Times, “On the basis of public announcements, Nvidia would seem to be in the lead at the moment. That was my assessment before this latest round of news, so this has only reinforced that view.”
Phil Magney, founder and principal at Vision Systems Intelligence (VSI), agreed. "I can hardly think of a single OEM not working (or developing) on Nvidia DrivePX technology. This does not mean they will all go into production with Nvidia. It's just that OEMs cannot afford to not examine the Nvidia eco-system for AI-based safety and automation technology."
How fluid is the situation?
Of course, none of the announced partnerships is exclusive. More important, the effort to design highly automated vehicles is a battle scene with a cast of thousands — ranging from chip suppliers to tier ones, OEMs, component suppliers and software developers.
A big question is the fluidity of these partnership arrangements. How easily might a car OEM committed to one platform (such as Drive PX) switch to another (such as the one by Intel/Mobileye/BMW)?
Strategy Analytics’ Riches said, “A commitment to a certain platform does not mean that it will never change — but does show a high degree of confidence that it is the best available solution for the short- and medium-term.”
He added, “There is always a cost to change. Software will be optimized for a particular hardware architecture, and engineers will get used to and skilled with a certain ecosystem of tools.”
When asked how hard to switch platforms, VSI's Magney told EE Times, "There is no plug and play Automated Vehicle (AV) stack. Once an OEM has committed it is probably going to stick with it, at least for that generation of vehicle."
VSI is currently engaged in developing an automated vehicle for its own research purposes. Magney said, "The development of AV functions is very difficult, as we are learning for ourselves. Stitching together all the code bases, synchronizing sensors, calibrating torque signals, controlling latencies, etc. takes massive investments in engineering resources."
He added, "Furthermore, developing and integrating the software stack into the hardware platform is also tough and very time consuming. Some of the AV Stacks come with an abstraction layer for adapting software applications a little bit easier, but in the case of AV Development platforms, there are still lots of gaps."
IHS Markit’s Ambroggi believes the situation is more nuanced and possibly more complex. By breaking down challenges associated with highly automated vehicle design into two chunks — hardware and software aspects — he noted, “Clearly, Nvidia wants to provide a complete solution for its customers (i.e. SW+HW), but in reality, in the commercial deal, the package sold might be having different models.”
De Ambroggi laid out two potential scenarios. The first scenario is that tier ones and OEMs might take the complete solution from Nvidia (i.e. DrivePX platform with Xavier + Nvidia AI software packages). Or, “They might take the software and algorithms portion only, remaining as much as possible agnostic on the hardware side.” He explained that “in this way OEMs can even trade better on SoC price, and ensure second source.”
Of course, “however you look at it, in software there is a lot of the ‘added value’ and the ‘high margin,’” he added.
What’s going for Nvidia, though, in De Ambroggi’s opinion, is its influence on the data center and training infrastructure, which he compares to Intel’s. “So, they are in the best position to bundle an optimized solution for their customers including IT infrastructure.”
Complicating the robo-car platform race is that “this AI-box will need to take input from different sensors from different suppliers (which Nvidia doesn’t really control) and deliver information/action to other generic ECUs (which are also coming from different suppliers),” De Ambroggi observed.
In the end, “What is inside the box might not matter as long as performance matches the requirements of the OEMs,” De Ambroggi concluded. “Clearly a minimum of standardization in I/O has to be ensured, and I think OEMs/tier ones will need to care for it.”
AImotive replaced by Zenuity
Partner swapping is already happening in the transition research to production phase, as indicated by the Nvidia-Volvo deal.
For Volvo’s "Drive Me" project in Gothenburg, Sweden, Volvo used Nvidia’s platform along with software from AImotive (previously known as AdasWorks).
Inside look at Volvo's self-driving car (Source: Nvidia)
But under a new agreement for the production model, Volvo, now in a partnership with Autoliv to develop and manufacture automotive safety systems, will replace AImotive with Zenuity. The latter is a newly formed automotive software development joint venture equally owned by Volvo Cars and Autoliv to develop next-generation self-driving car technologies, according to the companies
Zenuity will provide Volvo with self-driving software. Autoliv will also sell this software to third party OEMs, using its established and broad sales, marketing and distribution network.
Meanwhile, the system developed jointly by ZF and Hella, and using Nvidia's Drive PX platform, will combine front cameras with radar and software to create technology meeting the Euro NCAP safety certification for so-called "Level 3" driving.
More specifically, ZF, one of the industry’s largest automotive suppliers, and HELLA, a leading tier one supplier of camera perception software and sensor technologies, will “provide customers with a complete self-driving system that integrates front camera units, as well as supporting software functions and radar systems,” the companies said.
ZF's ProAI self-driving system (Source: Nvidia)
Easy migration path
Nvidia’s Shapiro called the transition from development to the R&D phase “an easy migration path.” No reset is required and systems are code-compatible.
In the end, Nvidia’s edge, said Shapiro, is the “openness” of its architecture. He called it “a software-defined, open automotive platform, powered by an AI system.”
This openness allows tier ones to customize and write new software, add new features and functions, and update them. The chip architecture can run many different versions of software, Shapiro explained. “It’s open and scalable, thus letting automakers to stay ahead.”
Magney attributed the momentum building around Nvidia to what the company has already built -- namely, "nearly a complete AV stack for the development of AV technologies." Magney added, "They have all the hardware, libraries, and development tools so that OEMs, tier-ones and other third-party developers can build up their applications. They also have (in collaboration with tier-one partners) production-ready solutions for incremental ADAS and automation."
Game over for other players?
De Ambroggi agrees that Nvidia is best positioned now on the AI side. It has already reached a “critical mass” of engagement with several automotive players.
While pegging Nvidia and Intel as “obvious leaders” on the hardware side, he cautioned that the market scenario could change, as carmakers might seek a platform more hardware-agnostic.
He added, “Let's remember that AI platforms are not the only electronic system to drive/control the car.” From the ADAS/autonomous driving perspective, especially in the short and middle term, the landscape could shift, he explained. As a result, “There is still place for more traditional electronics as well as suppliers [i.e. Texas Instruments, NXP].” In his opinion, “Safety and AI need some further progression to come closer and speak same language, and system redundancy is required.”
— Junko Yoshida, Chief International Correspondent, EE Times