Nvidia officials spent much of last week’s developer conference talking about the promising future of self-driving cars and showing off the technology the GPU vendor is developing to accelerate innovation in the space.
At the company’s GPU Technology Conference (GTC) in San Jose, Calif., CEO Jen-Hsun Huang and other executives unveiled such new offerings as the Drive PX 2 development board—powered by two next-generation Tegra chips and two new Pascal GPUs, and offering up to eight teraflops of performance—and new information on technologies like the company’s DriveWorks platform, which was first announced in January at the 2016 Consumer Electronics Show (CES). Huang said the software platform will ship this spring.
There also will be support for high-end mapping solutions like HERE.
Officials also spoke about how the drive for autonomous cars not only will help reduce the number of deaths on the highways, but also how it will change the ways cars look and are built. Danny Shapiro, senior director of Nvidia’s automotive unit, said that everything from the shape of the car, to the materials they are made of, to how the inside compartment will be configured will change.
The topic of self-driving vehicles also permeated the show floor, with cars on display and software being demonstrated. One of the keynote speakers during the three-day GTC was Toyota Research Institute CEO Gill Pratt, who talked about that company’s efforts in research and innovation in the space.
Nvidia’s focus on the connected car market isn’t surprising. Company officials said last year that Nvidia was putting significant resources behind the push into the space, which is getting a lot of attention not only from most major vehicle manufacturers but also a broad range of tech vendors that are delving into everything from artificial intelligence (AI) and real-time mapping to deep learning and supercomputer technologies aimed at the market.
Such capabilities take massive amounts of compute power—it essentially will take supercomputer performance—and Nvidia officials see the company’s GPUs as the products that will provide that power. The Drive PX 2 will come with not only the two Tegra chips and two Pascal GPUs, but also those GPUs will include custom acceleration for deep learning and up to 80GB per second of bandwidth.
“AI is coming to cars,” Huang said during his keynote address April 5. “Supercomputing is coming to cars.”
The DriveWorks development platform will include software that can work with up to 12 cameras and sensors that can gather information from all around the vehicle. Those sensors feed the data back into the platform that is armed with software that can process the data to give a 360-degree view of the environment around the car. Using machine-learning technologies, that data can then be used to help with vehicle operation, from control to navigation.
The sensor data also can be uploaded into the cloud to be used in a crowdsourcing method that can give other drivers close to real-time information on traffic and routes as well as aid in training self-driving cars. The DriveWorks platform supports such efforts as HERE.
ABI Research analysts expect a continued push for the use of information from vehicle safety features to be used for real-time mapping, particularly as cars become more autonomous.
“Crowdsourcing is crucial,” Dominique Bonte, managing director and vice president at ABI, said in a statement. “As connected vehicles include more low-cost, high-resolution sensors, cars will capture and upload this data to a central, cloud-based repository so that automotive companies, such as HERE, can crowdsource the information to build highly accurate, real-time precision maps. This is fueled by the rapid adoption of a wide range of active safety systems with more than 94 million longitudinal assistance ADAS [advanced driver assistance systems] expected to ship in 2026.”
Nvidia Puts the Focus on Self-Driving Cars at GTC
Included in DriveWorks will be a deep neural network that will aid in training autonomous cars in learning how to drive. Huang during his keynote showed how quickly—after 3,000 miles of driving supervised by a human—a car using Nvidia’s DaveNet network could drive on everything from highways to local roads, and over dirt roads and in the rain.
Huang also said that Nvidia’s technology will be at the center of the new “Roborace” series featuring driverless autonomous cars. Each of the 10 teams will have two identical cars, with the teams differentiating on the software that runs the cars. The race series will begin later this year.
As company officials outlined the Drive PX 2 and other offerings, others were talking about what the move to driverless cars will mean for the auto industry. According to Nvidia’s Shapiro, it could change what cars are made of. If there are fewer accidents on the roads, then cars may be able to be made from lighter materials, without the need for steel-reinforced frames or airbags. Without people driving, the inside of the car may feature swivel chairs so passengers can face each other, he said during a question-and-answer session after Huang’s keynote.
Shapiro also noted that these connected cars will be programmable through software, which means they can be updated throughout the life of the vehicle. Cars that initially may not have some features available on day one will be able to get those features later through updates.
Such capabilities will be important for the car industry and its customers, according to Bob O’Donnell, principal analyst with TECHnalysis Research. Upgradable cars could boost revenues in the industry because consumers will be willing to pay more for cars that can be updated via software and the cloud, O’Donnell told eWEEK. Rather than having a vehicle that automatically depreciates, drivers will have cars that will offer new features throughout the lifecycle.
“It’s time for upgradable cars,” he said. “It will give people the sense of increasing value in their cars.”
Also at GTC, Toyota’s Pratt talked about the automaker’s simulation work around autonomous vehicle technologies that is powered by Nvidia’s GPUs.