Nvidia, Intel take two paths to process autonomous driving

LAS VEGAS — As manufacturers begin the gritty development work to bring autonomous vehicle concepts to market, two paths are emerging to process the vast amounts of artificial intelligence needed to take the human out of the driver's seat.

In press conferences that took place one night apart, chipmakers Nvidia and Intel outlined their platforms to power automated driving. Nvidia's approach was holistic, processing as much information as efficiently as possible, while Intel was more conservative, limiting the situations in which to use processing power. While the industry is still years away from determining a best practice for such functions, experts say this technology is necessary to move forward, and could define long-term supplier agreements.

"AI is, in my opinion, a must-have when targeting advanced levels of autonomy," said Luca De Ambroggi, an analyst at IHS Markit. "It is not just appealing; performance requirements cannot be achieved just with traditional hand-coded algorithms."

Artificial intelligence lets vehicles learn from experience and make decisions on its own. The development of the technology has fast-forwarded autonomous vehicle development. However, extensive algorithms require significantly more computing power than current vehicles can handle.

Ultimately, the company that can achieve the safest yet most cost-efficient strategy will win customer contracts.

Nvidia's processors are popular in a variety of industries, from video games to health care, because they can handle enormous loads of information instantaneously. The company's approach to self-driving is no different.

"For self-driving, every single possibility needs to be considered," said Nvidia CEO Jensen Huang during the company's CES press conference. "The more computational capability you have, the more sensor backups you have, the better off you're going to be."

Late last year, Nvidia introduced its robotaxi platform, Pegasus, which Huang said will allow cars to consider every possible situation before making a decision.

The next night, CEO Brian Krzanich detailed Intel's self-driving platform developed with Mobileye, the camera and computer vision company Intel acquired last year. Intel said it developed the platform, which can power Level 4 and Level 5 self-driving, with production scale in mind.

The platform relies primarily on Mobileye cameras — which are already production-grade, compared with still-developing lidar technology — and makes decisions based on a long-term goal, such as an end destination, rather than evaluating every possible scenario.

"We use AI when necessary, but don't cram it in if it's not necessary," said Jack Weast, Intel's chief architect of autonomous driving solutions, in an interview. "We want it to be economically scalable, low-cost, low-power and scale to millions of vehicles."

Weighing the merits of the two approaches is complicated considering carmakers and suppliers are unclear on what final, production-ready self-driving cars' demands will be.

"Nobody, at the moment, knows what is needed for Level 4 [autonomous vehicles]," said Oliver Briemle, an autonomous engineer for supplier ZF. "Level 5 is even more expensive."

Shiraz Ahmed contributed to this report.