Self-Driving Cars Might Need Standards, but Whose?

The PC revolution, the internet boom, the smartphone economy — all were propelled along by a common set of technological standards. So will a standard platform or operating system be necessary to get autonomous cars rolling, too?

“There is certainly no doubt that many carmakers have expressed the idea that a more standardized platform would be attractive,” said John Wall, a senior vice president at QNX, which develops software that is used in millions of vehicles today.

Car companies have come to realize that existing in-car systems, with their tangled layers of software and morass of hundreds of embedded processors, are slow and overly complex by today’s computing standards. If autonomous cars are ever to become a reality, they will require even more powerful, more fully integrated and upgradable computing platforms, designers say.

“The feedback we hear is that the number of chips, the amount of wire — an Audi has over six miles of wire in it — is excessive,” said Danny Shapiro, senior director of automotive at the chip maker Nvidia, which is promoting its own autonomous driving platform.

A more centralized approach could bring huge cost savings by reducing the number of processors used in a car and thus reducing the need for the cabling among those processors. Less wiring would also lower a car’s weight to improve fuel economy, Mr. Shapiro said.

Nvidia is currently working with several automakers, including Audi, which plans to have its first self-driving vehicles on the road in 2020.

A common automotive platform could also make it easier to develop new software and safety systems for self-driving cars. “The goal is to reduce the cost of bringing innovation to the mass market,” said Eric Montague, a senior director at the speech recognition company Nuance.

Graphic

When Cars Drive Themselves

There are increasing signs that autonomous cars have arrived — and may be driving our city streets sooner than we think.

OPEN Graphic

Mr. Montague said the challenge today is that every car has a unique constellation of electronic control units, sensors and microphones. The software must be specially tailored to each model.

There is no shortage of companies vying to become the equivalent for self-driving cars of the Windows/Intel standard for PCs. Chip makers like Nvidia — as well as Intel — hope to establish their hardware as the new brains behind autonomous vehicles. Delphi, QNX and Waymo (the former Google self-driving project) are looking to put their software front and center as the operating system of choice. Each wants to sell its own car platform to automakers.

Some auto companies are embracing new partnerships. BMW is working with Intel, Mobileye and Here to bring self-driving cars to the road and plans to start testing such vehicles this year. Ford is working closely with QNX, a BlackBerry subsidiary, on the software needed for the vehicles it is testing. Fiat Chrysler has built a new experimental fleet of minivans for Waymo.

The traditional automotive culture, however, has emphasized independence. And some industry experts point to economic factors that argue against settling on a standard platform. Being dependent on a single vendor can mean higher prices.

“Automakers don’t want to rely on Apple or Google,” said Glen DeVos, vice president for software and services at Delphi. “But they also don’t want to reinvent the wheel all the time.”

Emblematic of that carmaker culture is Hyundai, which is mostly going it alone in working on its self-driving entry. The company recently demonstrated how far it has come, ferrying journalists in Las Vegas around in a self-driving version of its Ioniq electric compact car. Hyundai, which likes to emphasize that it even makes its own steel, has developed its own platform, striving for affordability.

“You don’t have to use a common system,” said Cason Grover, senior group manager for vehicle technology planning innovation at Hyundai. “And we don’t want to go down one path that hinders us in the future from introducing new innovations.”

Bosch, which supplies safety and technology systems to automakers, including its own self-parking technology, acknowledges that a single, one-size-fits-all system would be easier to work with. “But it would be a killer for innovation,” said Detlef Zerfowski, vice president for automotive system integration at Bosch.

Douglas L. Davis of Intel, recently charged with spearheading that company’s new self-driving car platform, Go, said, “Given the amount of computing power necessary for autonomous driving, we think it can benefit from greater commonality and predictable interfaces.”

“Mobileye already has the computer vision, for example,” he said. “So if the technology is good, why re-engineer it and take two to three years to get it into a product?”

Which design philosophy one chooses can have significant repercussions when it comes to the security of robotic cars, evoking visions of hackers causing mayhem by crashing cars and trucks into each other at highway speeds. It’s a problem engineers are keenly aware of.

“Having a common platform could have a downside,” Mr. Wall at QNX acknowledged. “If there’s a vulnerability in one car, it could mean there’s a vulnerability in every car.”

Conversely, Mr. Wall pointed out that de facto software standards mean that more programmers are focused on making those systems as secure as possible, versus having only a few coders working on a single, narrowly used program.

But by developing as much of the technology as possible themselves, some automakers believe they can solve problems more quickly.

“A side benefit of controlling the entire thing,” said Mr. Grover at Hyundai, “is that we know everything that’s going on.”

A lack of standards and a diversity of self-driving systems does present one other obvious challenge: Variety is not a virtue when cars must interact predictably with human drivers — and other robotic vehicles — to guarantee safety.

Riding in Hyundai’s self-driving Ioniq, for example, is like taking a Sunday drive with your grandmother. The car is extremely adept at staying squarely in its lane without ping-ponging back and forth, but it is also cautious in the extreme, stopping nine feet short of crosswalks and stubbornly refusing to go forward if a pedestrian looks poised to step off the sidewalk. It is behavior that can ignite road rage in nearby human drivers.

By comparison, Delphi’s test car, which uses an Intel computing platform installed in an Audi Q5, is more aggressive. It can easily merge into highway traffic and negotiate complex intersections. However, it treats pedestrians with less deference, taking right-hand corners more quickly — even though pedestrians may be contemplating entering the crosswalk.

“There’s a big difference in the algorithms and even sensor choices being used,” Mr. Grover said, which can affect how collision-prevention and lane-keeping systems work. When autonomous cars do hit the road, he said, you could have a fair amount of variation in how they behave.

“There are quite a few different approaches to creating autonomous systems,” said Ralf Herrtwich of Here, which is making high-resolution maps for the next generation of vehicles. Mr. Herrtwich said the necessary standards might come not from the car platform itself, though, but from the communications side.

In order to avoid accidents down the road and conditions like black ice that a car’s sensors cannot detect, self-driving vehicles will have to share information with each other, through so-called V2V or vehicle-to-vehicle communication. The National Highway Traffic Safety Administration has proposed that V2V equipment be installed in all cars in the future.

Indeed, there are already cars like the 2017 BMW 5 Series that can broadcast information about road hazards over a cellular connection, but only other BMWs can make use of the warnings.

Technology suppliers, platform vendors and automakers acknowledge that self-driving cars are still in their infancy and that as the vehicles mature, standards will be forthcoming.

Most car manufacturers are still struggling just to establish standards across their own models and brands, Mr. DeVos of Delphi said.

“Once that happens,” he said, “then we can talk about how to work across different companies.”

https://www.nytimes.com/2017/02/23/automobiles/wheels/self-driving-cars-standards.html?WT.mc_id=SmartBriefs-Newsletter&WT.mc_ev=click&ad-keywords=smartbriefsnl&_r=0