A new pragmatic focus on helping drivers, not replacing them

LAS VEGAS - Volkswagen isn't trying to put hundreds of thousands of self-driving Golfs on the road this year. Its ambitions are more realistic: Start deploying automated commercial vehicles for specific uses in limited areas half a decade from now.

The plans for VW's new autonomous driving unit, announced last week, reflected a growing sense of pragmatism in the air at this year's CES event here. Gone for now are the promises of fully self-driving robotaxis being commonplace in 2020. Expectations for Level 4 autonomy - still a six-figure-per-vehicle proposition - have been reset. The hottest trend now might be widespread adoption of advanced driver-assist systems that don't let the driver tune out.

Asked whether people were overly optimistic about the rapid deployment of ubiquitous autonomous cars, Alex Hitzinger, CEO of the VW subsidiaries tasked with bringing autonomous technology to scale, didn't hesitate: "Absolutely," he said. "This is one of the hardest problems we have right now in the world. It is like going to Mars."

He calls the commercialization of autonomous driving "the mother of all systems- engineering problems."

Here's why: Basically, three expensive groups of ingredients are needed to bake up a Level 4 system that can function without human involvement in a defined geography under certain weather conditions or other limitations, he explained. Sensors must perceive the environment; the computing hardware must process the inputs like a brain; and the software must provide the knowledge to translate all the inputs into useful information and decide what to do with it.

The type and placement of the sensors determine what algorithms are needed, and the nature of the software determines what hardware is needed to execute the programs.

If the sensor strategy is changed - say, by no longer using lidar, or using more lidar, or rearranging the camera placement - then the algorithms change, which alters the demands on the hardware.

"So you see, it's a circular systems-engineering optimization effort," he said. "And that all needs to be combined with the overall vehicle. It's an extremely complex topic."

Given the low volume of current pilot programs, a reliable Level 4 system costs upward of $100,000 per vehicle — even as much as $300,000, he said. By working through that circular optimization, he said, the price may fall to more like $10,000 by 2030.

Somewhere in the middle of this decade, it should become economically feasible to put such a system in a commercial vehicle for use in limited venues, such as a farm, a mine or a corporate campus.

And as potential uses grow, the spending can be spread more widely, driving down expenses and opening up more uses, which will add more scale.

This timeline — and deployment strategy — is very similar at ZF Friedrichshafen, the privately owned German supplier that is acquiring truck-braking specialist Wabco.

Among its business lines, ZF makes components needed for automated driving systems. But its immediate growth opportunities are in supplying parts or entire systems for advanced driver-assist systems that take some burdens away from drivers but leave them responsible for the vehicle’s safe operation.

Last week at CES, ZF announced a contract with a major Asian automaker to supply a full Level 2 system for much less than $1,000 per vehicle.

The system features not only adaptive cruise control and lane-keeping assist, but lane-centering, automated lane-changing, traffic-jam assist and even overtaking of slower-moving vehicles.

ZF calls it “Level 2+” to indicate that it is an advanced system but still short of letting the driver disengage. The widely used automation definitions set by SAE International don’t designate gradations within the levels.

ZF CEO Wolf-Henning Scheider said he doesn’t see Level 3 — where the computer can drive the car but the human has to be ready to take control — as one that can work in the real world. If the driver is allowed to go to sleep, for instance, it may take 30 to 60 seconds to effectively put the driver back in control of the vehicle. But for the car to drive for half a minute under duress, it essentially must have the capabilities of a Level 4 system that drives without human input, he said.

“In terms of passenger cars, we’re convinced that integrated Level 2+ systems are the right path to follow,” he told reporters at CES. “First of all, because there is currently no uniform legal framework for public-road operations on Level 3 and higher in any major market. Secondly, due to the associated high cost for higher-level automatic driving systems, we see no viable business case for individual car ownership.”

While declining volumes across the world’s biggest markets have spurred a “new pragmatism” toward self-driving cars, CES 2020 was still a Vegas show — glitzy and full of surprises. Hyundai showed an Uber air taxi concept, while Toyota announced ambitious plans to literally build a city of the future at the foot of Mount Fuji. And Sony, the Japanese entertainment and consumer-tech giant, even unveiled a sensor-laden concept car built with German supplier Benteler Automotive.

But aside from the handful of fanciful concepts, much of the discussion was around shorter-term solutions to more mundane problems, such as congested urban highways and clean-air regulations. Honda highlighted startups it has cultivated through its Xcelerator program, including one that makes lightweight exoskeletons to reduce strain on factory workers.

More modest steps can have the advantage of preparing society for some of the more transformative changes that may come later. ZF’s Scheider suggested that advanced driver-assist systems will “allow drivers to get used to semi-automatic functions in a safe manner.”

The more gradual progression makes sense, particularly while automakers face daunting costs to meet regulatory demands for electrification, said Karl Brauer, executive publisher for Cox Automotive.

“It feels like we’re settling into a world where we’ll prudently move toward full self-driving without the rush of anticipation of being on the verge of self-driving cars,” he said in an email. “I think the anticipation of autonomous technology has been replaced by the anticipation of more widespread EV adoption.”

And with the number of electric vehicles about to hit the market, he added, “that seems like a more realistic — though possibly still a bit too optimistic — sense of where the industry is heading in the near term.”

Ambarella, a smaller publicly traded supplier of computer chips designed to manage visual inputs, demonstrated a car that drove a preset route at night without human input, in part to show that a camera-and-radar system — without lidar — could see well enough at night to drive safely.

But its main news for the show was a pair of systems-on-chip that can manage inputs from multiple cameras, including one to monitor the driver, that operate efficiently so they won’t drain an EV’s battery.

“You need a lot of processing power on the chip — and that’s where we come in,” Senya Pertsel, Ambarella’s senior director of marketing, who is responsible for the company’s automotive business, said in a telephone interview.

After several years of broken promises from the industry about fully automated systems being just around the corner in 2017 and 2018, now “we see a winding road for that true autonomous driving” for passenger cars, said Aaron Jefferson, ZF’s director of product planning and strategy. The company is committed to figuring it out for commercial vehicles, people movers and cargo, but it will take more time, he said, because “the use case is difficult. The cost is difficult to justify, which is why we’re focused on a Level 2+” system.