Driverless cars’ bigger challenge
Getting the technology right is only the first step. Mass deployment will pose major hurdles.
Driving down the eight hairpin turns of San Francisco’s touristfilled Lombard Street in a futuristic self-driving car sounds sexy, adventurous, even terrifying. But in a Zoox vehicle, the experience is remarkably dull, like being piloted by a well-practiced taxi driver trying to impress a safety commissioner.
Boring, of course, is a big achievement — a testament to how Zoox, a self-driving start-up launched in 2014, has made quick progress by tackling the most complicated urban settings.
The Toyota Highlander, fitted with Zoox hardware and software, performs each turn with a curious mathematical precision. For now, two safety drivers are present, but they sit idle even as the vehicle navigates through a six-way unprotected intersection.
“We’ve gone out of our way to pick the most challenging ‘geofences,’ ” says co-founder Jesse Levinson, referring to confined areas. “And even within those, we put extra attention on the most challenging routes because you just learn so much more.”
The upshot is that the technological challenge behind self-driving cars is in its final stages. The next step should be straightforward: mass deployment.
At least, that is what the industry has long assumed. Since Google launched its “moon shot project” a decade ago to create a fully driverless vehicle, the engineering challenge of making a robot drive more safely than a human has been the focus. But now the battleground in autonomous cars is shifting.
There is a growing recognition that getting the self-driving algorithms right is merely an entry ticket for the much bigger challenge of commercialization — a test that requires government approval, public trust, brand marketing, the ability to manufacture at scale and the technical know-how to manage a fleet that competes with the likes of Uber and Lyft on timely pickups.
Numerous experts believe the industry is in for a reckoning. Amnon Shashua, chief executive of Mobileye, the driver-assist leader acquired by Intel for $15 billion in 2017, dismisses small-scale autonomous demonstrations as a “science project” with little relevance to deploying a viable robotaxi fleet.
“You can do nice demos, but bringing things to a level of productization — the level of safety that you need is completely out of the scope of many of the actors,” he says. Shaun Stewart, former chief business development officer at Google sibling Waymo, calls the engineering challenge “a huge feat — it’s more difficult than landing people on the moon.”
But, he adds: “It’s just one milestone. How do you succeed in the commercial endeavors around the technology? It’s a completely different challenge and requires a completely different skill set and experience set.”
Cruise knows this lesson well. In
July, the San Francisco company postponed plans to deploy a f leet of driverless cars this year after executives realized that it is “probably 10,000 times harder” to carry out large-scale deployment than demonstrating how an autonomous car drives around the block, says Dan Ammann, Cruise’s president.
“It’s like we are at the 24 Hours of Le Mans [race] and everybody is piling into turn one in the first lap,” he adds.
James Farley, Ford’s president of new businesses, technology and strategy, says the industry has been “focused way too much” on the engineering challenge rather than studying how to build a consumer service business.
“The self-driving system is a core competency, it’s necessary — but it’s not sufficient to create a new business,” he told the Financial Times’ Future of the Car Summit in Detroit last month. “There is still so much uncertainty around the manufacturability of the sensing systems, the regulatory environment [and] the scaling of the customer-facing businesses.”
Until recently, the running assumption has been that whoever is earliest to build a driverless fleet will attain a first-mover advantage wherein every mile driven improves the algorithms, thus enhancing safety and brand image. The idea was that autonomous cars would be a “winner takes most” market, akin to how Google won search or Facebook dominated social media, creating the “network effect” that would solidify market share.
In 2016, Travis Kalanick, founder and then-chief executive of Uber, said that if a carmaker could produce a truly autonomous vehicle, he would buy half a million of them immediately. Being first, he reasoned, was existential.
His premise was that the economics of ride-hailing could all be worked out if two-thirds of the cost equation — the driver — were taken out. But the details of how this would work, what the costs of operating the vehicles would be, or where they might launch, were treated as simply an afterthought.
Scott Corwin, head of Deloitte’s future of mobility team, says the industry has over the last decade spent untold sums perfecting selfdriving systems like there is a “race to win the Willy Wonka golden ticket.” But it is becoming clear that getting the technology right is only a first step.
“There’s no way you could do a national rollout, in any sense, for this technology. The cost of capital would be bananas,” says Gartner analyst Mike Ramsey. “They have other challenges that are huge and unresolved: the business case, the government regulations. That will slow things down.”
The rollout of autonomous cars will still happen over the next decade but with “limited market launches in pretty controlled environments, where variables are pretty knowable,” Corwin says. Local laws and political cultures are likely to lead to a host of different operators in U.S. cities, while there are unique regulatory challenges in major markets such as China, Japan and Europe.
Zoox tests almost exclusively in the dense cities of San Francisco and Las Vegas — less so on the highways that connect them. That challenge is being solved by groups such as Pronto.AI and Torc Robotics, which want to aid long-haul trucking groups. Waymo has focused on suburban environments in Silicon Valley and Arizona, while the start-up Voyage has targeted retirement communities in Florida, where a wrong turn is more likely to kill one of the turkeys that roam the area than a cyclist.
The idea that expertise in selfdriving is the key differentiator is being turned on its head. BMW, Daimler, Ford, GM, Honda and
Volkswagen are all partnering with specialist autonomous car companies, suggesting that the engineering part of the equation is becoming commodified, whereas the service side is how each brand will set itself apart.
Ford’s Farley speaks of finding “the Model T” of mobility services. “The differentiating part of this business, after the tech gets stood up, is going to be that customer facing business,” he says.
The quest to build an autonomous car is often portrayed as a single race. But the emerging market will be fragmented by different business models.
Zoox is the only autonomy start-up seeking to master the triple challenge of creating a driverless system, building a ride-hailing network and manufacturing a bespoke autonomous vehicle at scale. As one rival put it, Zoox is betting on “one too many miracles.”
Executives say a key advantage is the design of their vehicle — not yet unveiled in public — which is electric, bidirectional like a train, and free of a steering wheel or pedals. It features carriage seating so passengers can face each other and a host of innovations so that the highest safety standards are met.
“All the ride hailing or ride sharing that put people in the back seats are actually putting you in the area that doesn’t have very high crashworthiness standards,” says Mark Rosekind, head of safety innovation at Zoox and former director of the National Highway Traffic Safety Administration.
“In our vehicle, with the carriage seating, we’re going to have the same high front seat-level standards for all seating positions. If you’re not building it from the ground up, it’d be almost impossible to be able to do that kind of thing.”
But Zoox’s big ambitions could spell trouble. To date it has raised about $1 billion in equity — an impressive figure for a 5-year-old company, but a tiny sum for capital-intensive carmakers.
Aurora, a younger autonomous driving start-up led by Chris Urmson, the former project lead at Waymo, has raised $690 million from the likes of Amazon and Sequoia. But this looks more adequate given its more focused ambition to build what it calls “the driver” — a plug-and-play hardware and software set that it plans to supply to carmakers including Hyundai and Kia.
Urmson says he and his cofounders “don’t have a clue” how to build a vehicle or manage a ridehailing network, nor do they particularly care. “We recognize there is an immense amount of noise in the space, so we like to talk about the things we understand.”
Waymo and Cruise, which have developed their hardware and software in-house, are backed by vast capital and have partnered with carmakers — Waymo with Fiat and Jaguar, Cruise with GM and Honda. They believe a holistic approach is best to meet the rigorous standards of safety and quality.
“What Uber and Lyft ... have done is disrupt the taxi business,” Cruise’s Ammann says. “What they haven’t done is solved safety, they haven’t solved congestion — it’s probably worse. They haven’t solved emissions, or getting other cars off the road. What we’re aiming to do is to actually create what is, in essence, a new mode of transportation that fixes all those underlying issues.”
Even with the help of big carmakers, that goal may prove too lofty, like expecting Boeing to not only build the plane but manage all the ticket desks and run the airline.
Gartner’s Ramsey says it’s more likely that Waymo and Cruise will eventually outsource operations to a specialized network or to public transit operators. “It will be exceedingly hard for any company to develop the technology and then roll it out to every city,” he says.
Meanwhile, ride-hailing groups argue that their existing networks will be a major advantage enabling them to deploy a “hybrid system” of human drivers coupled with robotaxis.
Lyft co-founder John Zimmer says driverless-only fleets are in an unenviable situation. If they choose to service peak hours of traffic, then they will have a bunch of zombie cars driving around, causing congestion as they await bookings. They can choose to service only average traffic, but that will frustrate users in peak hours. A smaller fleet of driverless vehicles, mixed with human drivers who work when they want to, does not face this conundrum.
“I don’t see how you can make a case that anyone could compete with us, or any network, on utilization,” he says.