The High-Stakes Race to Rid the World of Human Drivers
Google's new self-driving prototype car during a demonstration at the Google campus in Mountain View, Calif. Tony Avelar/AP File Photo
The competition is fierce, the key players are billionaires, but the path—and even the destination—remains uncertain.
The race to bring driverless cars to the masses is only just beginning, but already it is a fight for the ages. The competition is fierce, secretive, and elite. It pits Apple against Google against Tesla against Uber: all titans of Silicon Valley, in many ways as enigmatic as they are revered.
As these technology giants zero in on the car industry, global automakers are being forced to dramatically rethink what it means to build a vehicle for the first time in a century. Aspects of this race evoke several pivotal moments in technological history: the construction of railroads, the dawn of electric light, the birth of the automobile, the beginning of aviation. There’s no precedent for what engineers are trying to build now, and no single blueprint for how to build it.
Self-driving cars promise to create a new kind of leisure, offering passengers additional time for reading books, writing email, knitting, practicing an instrument, cracking open a beer, taking a catnap, and any number of other diversions. People who are unable to drive themselves could experience a new kind of independence. And self-driving cars could re-contextualize land-use on massive scales. In this imagined mobility utopia, drone trucks would haul packages across the country and no human would have to circle a city block in search of a parking spot.
If self-driving vehicles deliver on their promises, they will save millions of lives over the course of a few decades, destroy and create entire industries, and fundamentally change the human relationship with space and time. All of which is why some of the planet’s most valuable companies are pouring billions of dollars into the effort to build driverless cars.
“This is an arms race,” said Larry Burns, a professor of engineering at the University of Michigan and a former GM executive who also serves as an advisor to Google. “You’re going to see a new age for the automobile.”
* * *
Many people have declared Google the frontrunner in the race for self-driving cars. The company has the road experience, mapping databases, artificial intelligence know-how, and, presumably, a significant head start. As of October, its fleet of vehicles had logged 1.3 million miles of test-driving in fully-autonomous mode since 2012—the distance-equivalent of 90 years of human driving, the company said. But that doesn’t mean Google will ultimately win. The major players each bring unique and formidable advantages.
Apple has the software experience, the potential for integrating its popular devices in interesting ways, and a track record for upending major markets the way it did in the realms of digital music and mobile phones. (Oh, and oodles of money, even by Silicon Valley standards.) Tesla has experience making and selling cars, glowing safety reviews for its existing vehicles, and a devoted customer base.
Uber has robotics experts, reams of mapping data, a mind-boggling trove of information about where people go and when, plus a wildly popular service into which driverless cars could be seamlessly incorporated. The idea that self-driving vehicles could simultaneously drive up Uber’s profits and lower the cost of a ride to Uber’s passengers makes for clear financial incentive to remove humans from behind the wheel.
Put simply, Uber can be expensive because “you’re not just paying for the car, you’re paying for the other dude in the car... who’s driving,” the company’s CEO, Travis Kalanick said in an interview with Recode last year.
Among Silicon Valley’s most buzzed-about players in the self-driving car space, Uber and Apple are the most mysterious. Neither company would agree to interviews for this story, despite repeated requests over a period of months, but there have been clues about their respective plans.
“Both companies have hired people who are extremely strong in issues of automated vehicles,” said Chris Gerdes, a professor of mechanical engineering at Stanford University. “They've built up strong teams that, from what I can tell, are doing something. But I do not know what they are doing. There is a joke in the Bay Area that if you know somebody in the vehicle space and they left their old job but they haven't updated their LinkedIn, they must be at Apple.”
In September, The Wall Street Journal reported Apple has a target ship-date of 2019 for an electric vehicle, code-named Project Titan, and that making that electric car fully autonomous is “part of the product’s long-term plans.” That news came months after reports from The Guardian that Apple lawyers had taken meetings with officials at California’s Department of Motor Vehicles, and that engineers from the company visited an enormous high-security facility in the Bay Area that’s being used as a testing ground for self-driving vehicles.
Around the same time last winter, Uber stunned the academic world by reportedly poaching 40 robotics researchers from Carnegie Mellon University. Though the director of the school’s National Robotics Engineering Center stressed to the Pittsburgh Business Times that the center lost “far less than half” of its staff to Uber, several senior engineers were among those who left, and other media reports at the time characterized the center as “gutted” and “reeling.” (The school's dean of computer science, Andrew Moore, wouldn’t confirm the number of staff members who left for Uber, saying he couldn’t remember exactly who left and when. “I actually don’t want to rehash the details,” he told me in September.)
Last February, Uber announced the creation of a new research facility in Pittsburgh, dedicated to developing new technologies related to driverless cars and mapping. Kalanick told Recode last year, “I’m not going to be manufacturing cars. That’s not what I plan on doing. Somebody’s gotta make them.”
Since then, however, Uber's chief product officer, Jeff Holden, seems to have hedged a bit.
“It’s possible,” he told Recode earlier this year, addressing the question of whether Uber might build its own self-driving vehicle after all. “There’s a lot of possibilities.”
Traditional automakers, too, have scrambled to establish themselves as leaders in this space—though perhaps not quickly enough.
“Why didn’t an auto company do what Uber did? Why didn’t they hire those 40 people that Uber did?” said Burns, the University of Michigan professor and former GM executive. “I don’t think auto companies had a bone-deep belief that it was possible. I don’t think they had that belief because they weren’t developing the state-of the-art [technologies]. The only people who really know what’s possible with driverless vehicles are the ones working in that frontier of knowledge.”
* * *
Tech firms and automakers aren’t just flirting with autonomous technologies on test tracks and in lab environments. The race to make self-driving vehicles ubiquitous is already playing out on public roadways.
There’s proof of this on YouTube, where a flood of new videos from Tesla drivers began appearing as soon as the company’s latest Autopilot software was released in October. Every morning John Leonard, an engineering professor at the Massachusetts Institute of Technology, would search for the latest examples of people who’d shared footage of themselves testing out the car’s new feature. What he found, in many cases, was alarming.
“They’re not all being insanely stupid,” Leonard told me. “But some of these people are totally reckless.”
Footage shows one car swerving toward a guardrail, another veering into oncoming traffic, and yet another exiting the freeway to the surprise of the driver. In most cases, the drivers had their hands completely off the steering wheel until the unexpected happened. In one video, a man holds an opened newspaper in front of him, blocking the wheel and the windshield as the car drives. “This is meant to show the absurdity of how the system could be misused today,” the description of the video says.
“They advise you keep your hands on the steering wheel when using the auto-steer, but as we’re testing, you really don’t need to,” said another man in a video that’s had more than 1 million hits on YouTube. He sits with his arms folded across his chest, traveling 75 miles per hour on the highway. “No hands, no feet, and I’m not nervous at all, really,” he said.
Autopilot is still in beta, and Tesla says it is transmitting data from the people testing the software so engineers can make improvements. (“One of the most surprising things about Autopilot is that Tesla owners are willingly taking part in the research and development of a highly advanced system that takes over steering,” Consumer Reports wrote recently.) Tesla compares Autopilot to the sensors and programming that pilots use so airplanes can fly themselves when conditions are clear.
“The driver is still responsible for, and ultimately in control of, the car,” Tesla says on its website. But some drivers are clearly pushing the limits.
The videos, and the behaviors they capture, highlight one of the most daunting challenges facing the autonomous-vehicle movement. Engineers aren’t just tasked with getting the technology right—and that, in itself, remains an enormous challenge—they have to manage consumer’s expectations along the way, and they have to do so in a crowded and intensely competitive field.
The amount of money at stake is potentially unprecedented, and could add up to billions of dollars per year, maybe more.
“If a first-mover captures a 10 percent share of the three trillion miles [driven in the United States] per year and makes 10 cents per mile, then the annual profit is $30 billion which is on par with Apple and ExxonMobil in good years,” Burns said. “So, the potential is huge.”
“The intensity of interest is amazing,” said Leonard, the M.I.T. engineer. “The Wright brothers jump into my brain immediately, but maybe it’s kind of like a decentralized space race. Like Sputnik, but between the traditional car companies and their suppliers versus tech companies and their startups.”
* * *
As the race intensifies, contenders have to decide which direction to take. There are two leading philosophies about how to build a self-driving car, and across the technology and automaking sectors, there’s disagreement about whether these approaches are complementary or in tension with one another.
The first approach, favored by Tesla and by most legacy car companies, is based on the idea that vehicles ought to incrementally add driver-assisting technologies so that safety improves and humans eventually do less of the actual driving over time.
Those features include advanced cruise control that can steer in certain environments, and functionalities like the Mercedes-Benz S-Class “traffic jam assist,” which trains the car to automatically follow the pace of traffic ahead at speeds under 37 miles per hour. Eventually, the thinking goes, all those stepwise technologies will yield a fully autonomous vehicle.
The second approach, led by Google, is that driverless cars must be built as fully autonomous from the start. Those who favor this strategy, known as a “Level 4” approach to automation, say it makes the most sense to build a car that can perform all safety-critical driving functions so that the only thing a human has to do is enter the vehicle and tell it where to go.
A vehicle of this nature wouldn’t need a steering wheel or brakes. It could start with a button. The passengers could take a nap, read a book, and otherwise disengage from the act of getting from here to there. This is the image of a self-driving car that’s most popular in wider culture, but it’s not the car that most companies are focused on building first.
Although Google is working on developing a Level 4 vehicle, it treats the cars it is now testing on public roads in Mountain View, California, and in Austin, Texas, as Level 3—meaning humans are expected to be available to take control of the car with comfortable transition time. (The numeric scale comes from the National Highway Traffic Safety Administration's vehicle-automation guidelines.)
“Both sides have arguments. I wouldn’t rule out either,” said Gerdes, the Stanford engineer. “There is a fundamental change, though, when you say the human is no longer responsible and the vehicle is responsible.”
The question of which path to take to full autonomy, a ground-up approach or a more gradual semi-autonomous one, is at the center of many debates about the technology. A more pressing question in the short-term is this: How much does a person’s perception of the computer’s job make a difference?
“This intermediate area where it may not be clear—is the vehicle responsible, or am I responsible?—is a hazardous place,” Gerdes told me. “There’s room for confusion that could reduce safety instead of increasing it.”
There’s evidence from the aviation industry that humans aren’t very good at navigating this area of uncertainty. Automation can make airplanes safer, but it also requires pilots to know which systems are turned on, and when. (And to stay up-to-date on their skills in case systems fail.) Self-driving systems may pose a similar safety paradox.
In 2012, when Google enlisted volunteers to test its new self-driving feature on the freeway, it made people sign contracts promising to stay engaged, even as the car drove itself.
“We told them this was early-stage technology and that they should pay attention 100 percent of the time—they needed to be ready to take over driving at any moment,” Google wrote in a blog post. “They signed forms promising to do this, and they knew they’d be on camera.”
And yet people didn’t pay attention even then. (Google has noticed similar levels of distraction with the human drivers its self-driving cars have encountered on the roads, including, in one case, a person playing a trumpet while driving.) “We saw some silly behavior, including someone who turned around and searched the back seat for his laptop to charge his phone—while traveling 65 miles per hour down the freeway!”
For Google, that experience prompted a pivot from developing incremental driver-assistance technologies to focusing on building a fully-autonomous Level 4 vehicle. There is additional research that backs up this line of thinking. In one government-funded study, researchers with the Virginia Tech Transportation Institute found that partially-automated driving modes had a significant impact on drivers’s attention spans, including 26 percent longer off-road glances compared with times they were just using cruise control. (Most of the time, in either case, drivers weren’t glancing away for more than two seconds.)
But how much will people’s behaviors change as their level of confidence in semi-autonomous vehicles increases?
“People trust technology very quickly once they see it works,” Google wrote in its monthly report about the self-driving car project in October. “As a result, it’s difficult for them to dip in and out of the task of driving when they are encouraged to switch off and relax.”
The tuning-out that Google observed among volunteer test-drivers is evident, too, in YouTube videos of people trying out the new Tesla software. Tesla’s Autopilot would probably be classified as a Level 2 system under the NHSTA definition, although plenty of people appear to be using it as though it’s a Level 3. In Level 2 vehicles, at least two primary-control functions are automated in unison as a way to relieve the driver of control of those functions—but the person is still expected to pay attention to the road as normal. In a Level 3 vehicle, the driver can cede “full control of all safety-critical functions” but should still be available to take control occasionally.
“A situation where you think you are not driving, but you are, that is really dangerous,” said Marcus Rothoff, the director of Volvo’s Autonomous Driving Program in Sweden. “There isn’t a gray area. Either you are responsible or you are not responsible.”
The potential for even more uncertainty about who (or what) is in charge of driving functions as vehicles evolve has experts worried about a near-future in which cars with different levels of autonomy will share the roads.
“Even if the technology is ready for the public, the public is not necessarily ready for the technology,” Leonard told me. “People seem to really want to do stupid things.”
Leonard and others told me they believe this transitional period, and confusion over driver responsibility, could even thwart the development of fully self-driving cars entirely.
* * *
The timeline is uncertain, but that hasn’t stopped people from guessing when fully-autonomous cars might begin to take over the roads.
“When is hard to say,” Alonzo Kelly, a robotics professor at Carnegie Mellon, told me. “If your horizon is long enough, they’re inevitable.”
Two years ago, the CEO of Nissan predicted driverless cars would be in showrooms by 2020. Jaguar Land Rover says it wants to give people the option to use fully-autonomous driving modes in its cars in the next decade. Uber’s CEO, Kalanick, tweeted a casual estimate of 2030. Elon Musk has said he expects fully autonomous Teslas to be on the roads by 2023. Earlier this month, Musk tweeted that he’s personally interviewing and hiring new software engineers who can help Tesla Autopilot make the leap to full autonomy, calling the move “super high priority.”
Google remains among the most ambitious.
“My older son is 12 years old,” said Chris Urmson, the head of Google’s self-driving car program. “In four years, he gets a driver’s license. We don’t want him to have to.”
What Google has accomplished so far is impressive. No other company has tested autonomous vehicles on public roads the way it has. Google has worked hard, too, to cultivate a reputation for carrying out these tests safely. In each of the 16 accidents involving Google’s self-driving cars since 2012, the company says, a driverless vehicle has never been to blame.
Google is also, by far, the most transparent of the tech giants about some of its work on self-driving cars.
“It’s an interesting situation because one company has much greater insight into the depth of the problems and how close we are [to solving them] than anyone else,” said Gerdes, the Stanford professor. “They have better data than universities, they have better data than the government, they have better data than auto manufacturers, they have better data than any competing tech company.”
“Google is so far ahead of everyone else because the maps they use are so detailed and the LIDAR they’re using gives so much rich information,” said Leonard, the mechanical engineering professor at M.I.T. “What Tesla has is just a single camera, a single radar. Google’s sensor gives 1 million data points a second.” (Tesla would probably also point out the “12 long-range ultrasonic sensors positioned to sense 16 feet around the car in every direction at all speeds,” as it boasts on its website, but the company declined an interview request.)
It can be tempting to declare Google an early winner in the race for self-driving cars—one byproduct of Google’s transparency, of course, is devoted media attention. Leonard, in a lecture over the summer, praised the company’s “amazing research,” but also cautioned its technology is “over-hyped and poorly misunderstood.” Google itself is conservative about where it stands.
“There’s so much to do,” Urmson said. “Let’s see how it actually goes. “It’s obviously really hard to tell what’s actually happening other places,” he continued. “Uber, and who knows what Apple’s doing, all of the automotive companies ... Right now, we’re at a phase where we’re all trying to move the technology forward as quickly and thoughtfully as possible and then we’ll start to see what happens in the marketplace in a few years.”
Looking ahead, there may be hope for the incumbents in the auto industry yet. For one thing, building the first market-ready fully-autonomous driverless car is no guarantee of business success. Standardization will likely trump innovation, which means legacy carmakers may still be able to capture the market.
“The first to market may not win the billions or patents,” said Arthur Wheaton, a director at the Worker Institute at Cornell University. “The first to be able to make autonomous vehicles safe, reliable, and affordable—with seamless interface with GPS mapping—will likely reap billions in profits. First is not the key; it is becoming the standard in industry.”
Wheaton offers Henry Ford’s legendary success as an example: “Most Americans credit Henry Ford with the first automobile. That is completely false. What Ford did was market his car to the masses instead of the rich and had a laser focus on cost cutting and increasing production.” Another thing, and it’s a big thing, is that legacy car companies already know how to build and sell cars. They have supply-chain experience in auto manufacturing that the tech giants, with the exception of Tesla, simply do not have.
“So with Apple, they have money to develop the vehicle, but do they have the production experience to be able to do that?” Wheaton said. “They don’t even build their iPhones. They have glitches with their phone. Now imagine that phone going 80 miles down the highway.”
Which means tech companies may see opportunities for partnerships with legacy companies. Urmson has said before that Google may end up partnering with a carmaker that would manufacture a vehicle containing Google’s software, for instance.
Thinking about driverless cars in the context of the mobile-phone revolution is useful for another reason: If fully self-driving cars reach the market, it’ll be in the world that smartphones made possible; a world of GPS-enabled maps, voice-controlled robot assistants, constant connectivity, and everything on-demand.
And the leading players—Apple, Google, and Uber—are all vying for leadership in a new sector of this world, which they created. They know what it’s like to demolish the incumbent in an industry because they’ve all done it before.
But legacy automakers aren’t banking on teaming up with Silicon Valley anyway. They’re making their own moves toward self-driving cars. Many traditional carmakers are adding driver-assistance features incrementally, and they describe plans to work toward fully-autonomous vehicles eventually.
Until these features are actually available, though, It can be difficult to discern serious engineering efforts from talking-points made to capitalize on one of the most buzzed-about topics in Silicon Valley—and in Detroit, for that matter.
“Industry incumbents have some very important strategic decisions to make, and the biggest mistake they could make is to be caught without the knowledge, at least, of what’s possible,” Burns said. “The roadway hasn’t been disrupted for 130 years. Something’s going to give here in the next five to 10 years for sure.”
In November, Toyota announced it is pouring $1 billion into a new research and development company, headquartered in Silicon Valley and focused on artificial intelligence. Mary Barra, the CEO of General Motors, told USA Today in October that her company has “a lot of efforts that are confidential that we’re working on in all aspects of autonomous.” (Of Apple’s plans to enter the car market, Barra said, “it would be foolish for me not to assume that they’re going to.”)
GM also just unveiled its new Super Cruise feature, rivaling Tesla’s Autopilot, which adds steering and lane-centering to GM’s existing adaptive cruise-control systems in 2017 model-year Cadillacs.
“It’s intended to reduce the driver workload and add a second set of eyes that will help avoid collisions where you may not have responded quick enough on your own,” said Dave Leone, the executive chief engineer for Cadillac.
The technology works by triangulating GPS data, information from a camera that can detect brightly-painted lanes, and LIDAR imaging. GM is also working on a system to monitor a driver’s eye movement to make sure that drivers stay focused on the road even if their hands aren’t on the wheel.
“We will be monitoring your eyes, and if it detects you’re looking away, it will tell you you need to take over,” Leone told me. “If you're looking in your lap as opposed to the windshield, it will know that, and it will give you an alert.”
If drivers ignore that alert and stay unfocused on the road, the car will “disengage” and “make you take over,” he said. GM also has sensory warning signals so that the driver’s seat will vibrate to warn of a potential collision.
Leone says all this represents a step toward fully autonomous vehicles, but a small one. “We are going to walk carefully, not run,” he said, before taking a modest jab at Tesla. “We aren’t going to let our customers test it for us.” (If you see a Tesla on the road, he joked, “stay behind them.”)
Volvo’s Drive Me project, which is set to launch in 2017, will put a fleet of 100 driverless cars on the highways in Sweden. (Humans will have to drive them to the highways, and take over in inclement weather, according to CNET, but the cars will be designed so they can pull to the side of the road safely without intervention if a human can’t react in time.)
At the same time, Volvo is developing what it calls “highly autonomous” features to add to its existing cars. Volvo’s philosophy is that a dual approach—working on full autonomy and semi-autonomy at the same time—is a way of building trust with customers in an uncharted technological realm. It’s also a way to give them a variety of options, while still working within the cultural space the car occupies.
“The car very early became a symbol of freedom, and we have to maintain that level of freedom because it’s part of our lives,” said Rothoff, the director of Volvo’s Autonomous Driving Program. “No one really feels free when you are in a congested highway. With autonomous cars, we’re expanding the concept of freedom. It will offer you control of your own time, and open up the freedom to spend that time how you want.”
* * *
Major shifts in transportation technology tend to reconfigure human perceptions of time. The time zones still used in the United States were famously established in response to the railways that had stitched together the country.
“What changes must now occur, in our way of looking at things, in our notions!” wrote Heinrich Heine in 1843. “Even the elementary concepts of time and space have begun to vacillate. Space is killed by the railways, and we are left with time alone.”
Self-driving cars similarly promise to “annihilate space and time,” to borrow a phrase from the 1850s. Whereas railways seemed to shrink the space between departure and destination, driverless cars promise to create new windows of time for people who would otherwise be stuck staring at another car’s bumper in traffic. Municipalities could, theoretically, order fleets of self-driving cars and turn old parking lots into tree-filled parks, reducing the need for and eliminating the cost of individual car ownership.
The potential implications for people who can’t drive themselves—due to disability, age, fatigue, drunkenness, or other factors—are huge. And as driverless cars outnumber human-driven ones, and begin communicating with each other and smart traffic signals, there’s potential to eradicate traffic jams.
“You’re going to see the emergence of a new mobility ecosystem,” Burns said. “It’s not a question of whether. The real questions are who, how, and when.”
Self-driving cars won’t just drive people, but goods.
“The opportunity to apply this [technology] to moving stuff is a huge thing to think about,” Google’s Urmson said. “We haven't even scratched the surface.”
In October, Daimler tested one of its driverless semi-trucks on a public highway in Germany. Driverless, in this case, is “Highway Pilot” mode, similar to Tesla’s Autopilot. The 18-wheelers are also being tested on Nevada roadways. Daimler has said the trucks will be available for companies to purchase around 2025. Gerdes, the Stanford professor, is also the cofounder of a startup called Peloton, which uses radar and wireless communication so that platoons of trucks can link up on the highway to share safety data, enable features like automatic braking, and encourage fuel efficiency.
“These technologies are really already here,” Gerdes told me. “We’re not yet at a point where you can jump in a car at any Point A and tell it to get you to Point B, but I think we will inevitably get there.”
If self-driving vehicles are poised to revolutionize transportation, they will transform mobility at least as much as trains and airplanes before them. Only this time, success will mean altering the role of humans in an unprecedented way—removing them from the driver’s seat altogether. This change is what imbues the larger conversation about self-driving cars with an existential quality.
“Some of these questions go right to the heart of artificial intelligence,” Leonard said. “I don’t want to say ‘never,’ but it’s not a foregone conclusion that these things will be ubiquitous.”
“But there’s a chance,” said Urmson, at Google. “All these wonderful impacts it can have in the world.”
Today, it would be foolish not to be skeptical of self-driving cars and the utopia their devotees promise, but it would be inhuman not to try to build them anyway.