Nvidia is gaining partners as OEMs and developers are a little turned off by Mobileye and Intel’s closed approach to the automated vehicle platform, analysts say.
Toyota will be using Nvidia's Drive PX AI automotive platform to power advanced autonomous driving systems planned for market introduction, Nvidia has announced at its GPU Technology Conference.
Mike Demler, a senior analyst at The Linley Group, described Toyota’s move as “potentially a big deal.”
In the brewing battle between Nvidia’s AI car computing platform and an Intel-Mobileye platform, Nvidia now appears to be building momentum.
According to Egil Juliussen, director research, Infotainment & ADAS at IHS Automotive, Toyota has become the fourth major car OEM publicly committed to Nvidia’s Drive PX for their highly automated vehicle. The other three OEMs are Audi, Daimler and VW Group.
In addition to those OEMs—which include the world’s two biggest carmakers Toyota and VW, Juliussen added that Nvidia also previously picked up smaller OEMs including Volvo, Tesla and Nio (formerly known as NextEV). Since tier ones such as Bosch and ZF have also embraced Nvidia’s hardware platform, Juliussen believes that this “will probably help Nvidia getting other OEMs on board.”
Figure 1: In the brewing battle between Nvidia’s AI car computing platform and an Intel-Mobileye platform, Nvidia now appears to be building momentum.
Demler, who attended Nvidia’s conference Wednesday, also noted that Argo.ai, Ford’s autonomous driving group, gave a presentation on “Deep Learning in Argo.ai’s Autonomous Vehicles.”
Of course, it’s important to note that the automotive industry is “still in a very early stage of development for Level 4 and Level 5 self-driving cars,” cautioned Demler. It’s premature to declare any platform’s victory. Juliussen noted that “other platforms for autonomous driving are likely to appear.”
But so far, it’s hard to deny that Nvidia is picking up steam.
A year ago, Toyota Research Institute CEO Gill Pratt came to Nvidia’s conference to deliver a keynote speech, in which he emphasised why simulation is the key to autonomous driving. By leveraging Nvidia’s GPU-powered platform and developing simulation programs, Pratt explained that it’s incumbent upon researchers at the Toyota Research Institute to tackle “corner cases” that happen rarely during trillions of miles of driving in the real world.
Without simulations to augment learning from huge quantities of real-world data, miles of cumulative driving alone won’t help the industry find answers for such edge cases, he explained.
Fast-forward to May 10, 2017. “Engineering teams from the two companies—Toyota and Nvidia—are already developing sophisticated software on Nvidia’s AI platform,” Nvidia announced. The goal is to “enhance the capabilities of Toyota vehicles, enabling them to better understand the massive volume of data generated by sensors on the car and to handle the broad spectrum of autonomous driving situation,” according to Nvidia.
Phil Magney, founder and principal at VSI (Vision Systems Intelligence), told EE Times that among Nvidia's growing list of automotive partners, some are pilot programmess while others are in production. “In the case of Toyota this is a production deal to use the Drive PX (or elements of it) to improve automation and safety in future vehicles.”
Magney also added, “It is my impression that the Toyota deal is as much about safety as it is about automation.”
In his opinion, Toyota’s way of “using AI to improve the safety of the vehicle is actually pretty shrewd because AI can make inferences about potentially dangerous scenes.” He explained, “It is like having another set of eyes and another brain to interpret the infinite number of scenes and edge cases.”
Magney said, “Safety still sells cars and for the next couple of decades (before we reach critical mass for fleet-based automation) cars are going to have lots of tech including automation features and co-pilot technologies which will vastly improve safety.”
Not to be outdone by Nvidia, Intel’s Katy Winter, vice president and general manager of Automated Driving Group at Intel, issued a statement on Wednesday.
Reiterating the Intel-as-a-data company pitch, she discussed what it takes to process over 4TB of data per 90 minutes of driving. “Intel is the only company that offers a complete suite of solutions that OEMs and tier-one suppliers need to address this enormous data challenge. Intel's solutions are designed to work with this heterogeneous data not just inside the vehicle but throughout the network and across the cloud.”
Although Intel has not before detailed its highly automated vehicle architecture, Winter noted, “Our CPUs, FPGAs, AI platform and software solutions have already been fine-tuned to address the specific requirements of our automotive partners who are working to bring highly automated and fully autonomous vehicles to production.”
So far, Intel’s known partners include Mobileye, whose acquisition Intel announced earlier this year. The others are Baidu, BMW and Delphi. Hinting that there are many other deals she can’t talk about, Winter said, “We encourage you to look at many of the autonomous test cars running on the roads today, open the trunk and look for Intel Inside."
Asked why Nvidia appears to be gaining partners, many industry analysts pointed out that OEMs and developers are a little turned off by Mobileye/Intel’s closed approach to the automated vehicle platform.
The Linley Group’s Demler said, “We’re seeing tier ones and carmakers signing up to use Drive PX because it gives them a better development platform than Mobileye’s closed system.”
Noting that Intel’s Mobileye acquisition will take months to close, Demler said, “There currently is no competitor that can offer the same platform for development, training and inference.” He added, “It’s doubtful Mobileye-Intel will open up their platform to match CUDA-DNN, Drive PX, etc.”
Asked about the Intel/Mobileye/BMW platform, Magney said that its approach is similar to Nvidia’s, but the architectures are different.
He said, “Led by the processor companies, we are seeing a race to provide as much of the AV [automated vehicle] stack as possible. Rather than a node here or there, the big chip companies are all trying to assemble an eco-system of hardware, software and development tools. They are all offering up "platforms" with the hardware and software components, plus the development tools and simulations to build up the solutions.”
Nvidia’s lead in picking up more design wins can be also explained by the fact that it is seen as “the market leader in democratising AI,” Magney observed.
Deep learning and AI, however, aren’t a panacea for automated vehicles because scientists can’t explain how deep learning works, which makes it difficult for safety test engineering to validate the safety of autonomous cars.
Magney acknowledged that there is still some scepticism for AI used in automated vehicles, “because of the challenges associated with how to validate.”
He noted, “Nvidia carefully explained that while you may not be able to precisely pin point why an inference model does what it does, you can examine the layers to see what activates the network. You can isolate the problem, adjust the weights in the inference model and test it through simulation to check the outcomes.”
First published by EE Times U.S.