Intel, with Mobileye, streamlines autonomous vehicle decision-making

LAS VEGAS -- Intel and its recent acquisition, Mobileye, are streamlining the autonomous vehicle decision-making process and have entered new territory in high-definition mapping.

At a CES keynote speech here late Monday, Intel CEO Brian Krzanich detailed Intel's camera-based automated driving platform, which sets rules to make decisions and simplify processing, rather than evaluating every possible driving situation.

The company also announced partnerships with SAIC and NavInfo to develop autonomous vehicles and roadmaps in China.

“The rise of autonomous cars will be the most ambitious data project of our lifetime,” Krzanich said. “It will reshape our cities and transportation systems.”

Intel acquired Mobileye for $15 billion last year, giving it access to a wide range of automotive customers, computer vision and mapping technology. Less than a year into the marriage, the companies are beginning to show the fruits of their collaboration.

Simplified decision-making

Autonomous vehicles are able to make crucial driving decisions without human input by using deep learning -- a set of algorithms that can be trained to recognize and react to situations.

To simplify processing, the system identifies a long-term goal, such as arriving at a specified location, said Jack Weast, Intel's chief architect of autonomous driving solutions.

The computer ranks possible actions -- such as changing lanes or speeding up -- by their usefulness in reaching the vehicle’s longterm goal.

Each short-term decision is run through safety algorithms that set rules for variables such as speed, braking and following distance.

This approach allows the vehicle’s computer to consider a relatively small number of alternatives rather than conduct an inefficient “brute force” calculation of every possible route.

The platform, which uses two Mobileye EyeQ5 chips and an Intel Atom processor, is suitable for Level 3 vehicles -- which would include automated driving on highways -- up to Level 5, when no human supervision is needed.

The system relies primarily on camera data supplemented by lidar and radar, instead of a more costly lidar-first approach.

"This simplifies the amount of processing you need quite considerable," Weast said. "We use artificial intelligence when it's necessary, but don't cram it in there if it's not. It's a low-cost, low-power way to scale to millions of vehicles."

Chinese expansion

Krzanich also announced that Mobileye will expand its Road Experience Management service, which crowd-sources map data uploaded from vehicle cameras, to China through a partnership with mapmaker NavInfo and automaker SAIC.

With its huge population, China has been eyed as the ‘great white whale’ of digital mapping. The country has favored local companies over foreign tech giants, effectively barring Google Maps and Apple Maps.

As autonomous vehicles become reliant on high-definition maps to operate without a driver, mapping companies are scrambling to map roadways in every region. But China has remained out of reach for most.

Here, a mapmaker partially owned by BMW, Audi and Daimler, came close with a proposed investment from NavInfo and Chinese tech giant Tencent, but U.S. regulators blocked the deal in September.

Through its new partnership, Mobileye gains a major advantage in building a comprehensive crowd-sourced map. Intel said its Chinese partnership is a collaboration rather than an investment, which may allow it to avoid regulatory hurdles that other companies have experienced.

David Sedgwick contributed to this story.