The most startling thing about riding around in a driverless car is just how ordinary, even boring, it can be—once the shock passes of seeing a steering wheel move by itself. It takes about a minute for my brain to turn self-steering from stunning conjurer’s trick into something commonplace. That’s when I notice the other standout feature of driverless cars: everybody else on the road is passing us by. One vehicle zooms close up from behind, the classic sign of an impatient driver ticked off at the slowpoke in front. Once it becomes clear that the autonomous car won’t be intimidated by the tailgater into speeding up, the car behind swerves into the passing lane and zips by. A human driver would probably take this affront personally. The robot car? Not so much.
The reason why all the other cars are passing the Google driverless car—or rather, the self-driving car, to use the search company’s preferred nomenclature—is simple. It’s the only car on the road obeying the speed limit. And obeying the law is one of those things it does scrupulously, which will eventually raise an interesting safety issue down the line. Another thing it does—rather jarringly, given its usual non-daredevil driving style—is react like a cat.
I saw this behavioral change happen halfway into my ride, as we were driving down a tree-shaded street that cuts through the Google campus in Mountain View, California. All the buildings on this street house Google offices, labs, meeting rooms, and cafeterias where the food and coffee are shamelessly tasty, but the feel is more college campus than tech titan. Amid a steady stream of pedestrians and cyclists, the Google car pokes, smooth as butter, laser turret on top spinning and watching. I’m chatting with the test driver and her partner, who is tapping on a laptop displaying the stylized virtual reality of differently colored moving squares, circles, and vectors that represent the objects the car “sees.” Suddenly the robot driver jams on the brakes. Did I say driverless cars are boring? I look up and see a bearded fellow walking in front of the now stopped car. He is balancing his own full-sized laptop on one palm and forearm while typing away with the other hand, staring at the screen intently as he jaywalks right in front of the car. He finally looks up, vaguely surprised to see the Google’s Lexus SUV test car with the sensor array bolted on top, then gives a little wave before resuming his diagonal course across the street, back to typing and staring at his screen. Coders. They’re all over the campus, it seems, focused on task, oblivious to environment. I can’t help but think that a human driving the same street, maybe distracted by a conversation or taking a quick peek at a cell phone or just leaning over to tune the radio, could have been slower to react than the robot car, and the result could have been very different. Flying laptop. Flying nerd. Ambulance rides. Or worse.
Welcome to the brave new world of autonomous vehicles, where rules are followed, at least by the nonhumans, and where, if Google has its way, nobody will ever die in a car crash again.
Can that happen? Absolutely. Will it happen? That’s uncertain. But should it happen?
That’s my question to Ron Medford, the former number two at the National Highway Traffic Safety Administration, now safety director for the Google Self-Driving Car Project. We speak after my ride, after the requisite marveling at the gee-whiz technology, after Google has shown off what this car can (and, just as important, won’t) do. Can and should are very different things, after all, and Medford has been a car guy—a human-driven car guy—all his life. So I want to know: Why does he think we should all follow this path Google is pursuing, this odd yet compelling deviation from its core Internet-search gravy train?
Medford’s response is adamant that we absolutely should. His explanation boils down to two points: If you (1) care about your kids, your spouse, your parents, your siblings, the elderly, the infirm, or the lives of innocents, and (2) if you ever saw a human driver do something stupid, dangerous, or deadly (which is to say, if you’ve spent more than five minutes on the road), then you want self-driving cars to happen.
“As soon as possible,” he says. “As fast as we can.”
Most major carmakers are working on driverless technology, along with Google, a few smaller car technology companies, the major automotive parts makers Continental, Bosch, Delphi, and Tesla (where chairman Elon Musk predicts that human driving will be illegal someday), the rideshare leader Uber, and possibly Apple, although that company, as usual, isn’t talking. Almost all the automotive powers have large outposts in Silicon Valley now, a clear indication that the future of mobility will be guided by software, not hardware. Behind this extraordinary pursuit of new technology, two very different philosophies have emerged, one evolutionary and more comfortable for consumers and car lovers to accept, the other revolutionary and more disruptive, but with the bigger potential payoff.
The mainline carmakers’ approach is to treat autonomy as a feature, something to be introduced gradually without eliminating the standard car-owning and driving experience. Their autonomous cars leave drivers very much in the loop. First there’s highly adaptive cruise control now on the market that can automatically respond to highway conditions and brake to avoid collisions automatically—when drivers decide to switch it on (studies show many don’t). Then there’s parking assist, a real blessing for the parallel-parking challenged, one of the main reasons new drivers fail their driving tests (leading at least a dozen states to drop parallel parking from their tests). These technologies are already available in some cars. Next might be lane-following technology, so you could turn over control to the car while on a free-flowing freeway, but not in stop-and-go city traffic. It would be like a traditional car with an autopilot function, with the human driver expected to remain vigilant and ready to take over at any time. In fall 2015, Tesla transmitted an over-the-air update to its 2014 and later models that enabled such an autopilot feature, with warnings that the technology was still in its early phases and so drivers should keep their hands on the wheel and be prepared to take over at any time. The user reviews were mixed, because a system that requires constant supervision from the driver is more curiosity than useful tool. But it did hint at the potential of such technology.
Eventually, full autonomy on all types of streets and roads could be introduced by the big consumer carmakers, but with redundant controls for the human driver to use whenever he or she felt like driving. The closest analog would be jetliners with autopilots doing ordinary tasks but with human pilots supervising and in control at critical moments. Such cars are being tested in Europe, the U.S., and Asia. Volvo is using a ring road in Sweden’s second largest city, Gothenberg, where it expects to test one hundred cars in this mold with volunteer car owners by 2017. The University of Michigan has built a thirty-two-acre fake city to use as a test bed for the driverless products of Ford, General Motors, Honda, Nissan, and Delphi for safely testing both city and highway driving by robots. And Bosch is road-testing something it calls the “highly automated” vehicle, which is driven manually to a highway, where the driver can either continue at the wheel or switch to autopilot. When the car reaches the exit, it asks the driver to resume control; if there is no response, the car pulls over and stops. This approach has another advantage: automated driving on limited-access freeways with divided traffic lanes, off-ramps instead of intersections, and no traffic signals poses a significantly simpler technical challenge than surface street driving. That level of autonomous technology is virtually off-the-shelf at this point, as Tesla proved with what amounted to a customer beta test.
Carmakers plan to pitch such systems as the best of both worlds, a car, much like today’s cars, with a really cool set of high-tech options. These options might improve safety when engaged, make drive time more relaxed and productive, and get people used to the idea of automation without the potentially off-putting discomfort about giving up total control to a machine. No one would—or could—be forced into giving up driving in these scenarios. We’d still enjoy the “freedom” of driving—as well as the freedom of driving drunk, distracted, or over the speed limit. In this scenario, while it might be true that such technology will take drivers out of the loop someday, it would be many years in the future, so there’s little point in pushing hard now. With 265 million cars, pickups, and other passenger vehicles on the road in the U.S., nothing changes overnight.
The Google approach is different: the company’s goal is to perfect the technology, then bring it to market whole hog, with completely autonomous cars that don’t even have steering wheels or gas pedals. Just a start/stop button and an interface to tell the car the desired destination. The faster that transition happens, Google suggests, the better off the world will be. They started with stock Lexus SUVs as test beds for the autonomous technology, but in mid–2015, they began road-testing their own design for little electric pod cars designed from the ground up for autonomous city driving. This approach dives into the discomfort of ceding control to a computer and pitches it as a new freedom: freedom from having to pay attention. So you want to text in your car? Get distracted? Doze off? Feel free! A major cause of death and destruction in regular cars—distraction—is a welcome feature in Google’s driverless reality.
“We think ours is the correct approach,” explains Medford, the former number two at the National Highway Traffic Safety Administration before coming to Google. The handoff from robot to human is potentially the most dangerous moment when cars are partially automated, he says, particularly if a person who has not been paying attention to driving conditions is asked to take control in an emergency, as some of the early partially automated systems require. Google is more interested in reinventing driving than in adding features to existing cars. If safety is the overriding goal, along with helping the disabled and the elderly who can’t drive become mobile, then, Medford says, Google’s way is the right way.
The safety gains of full and ubiquitous automation are clear. Robots don’t drink and drive, or get distracted, or get drowsy at the wheel, or speed, or randomly cross the centerline, or blow through stop signs or red lights. The main causes of traffic death and destruction simply go away. Robots’ reaction time is close to instantaneous. Their radars and lasers have pinpoint accuracy and allow them to “see” 360 degrees simultaneously, through hedges and leaves, to identify pedestrians approaching crosswalks. The most powerful autonomous car sensor is the lidar, which sounds like a type of radar but is a completely different animal. The spinning scanner emits bursts of illumination at a rate of nearly a million flashes per second, a flickering light invisible to human eyes that bounces off everything in the environment—people, cars, buildings, curbs, squirrels. The process is something like insanely fast photography, but the result is a kind of three-dimensional machine vision that allows the car to perceive its environment in incredible detail in real time, with measurement of objects and their height, width, depth, and distance accurate to the millimeter. Lidar makes it possible for the driverless car to calculate a collision course with a wayward oncoming vehicle when it is still comfortably distant, and shift course so smoothly that no one even notices. Nothing can sneak up on the Google Car. In a world full of human drivers doing stupid things, autonomous cars are the ultimate defensive drivers.
But when it’s all robots on the road, then the need to defend lessens and the real magic begins. When driving with their autonomous brethren, these cars are capable of cruising bumper to bumper on the narrowest of lanes at high speeds without a hiccup, driving in tight formation like precision aircraft teams, never wavering. It’s as if they are on invisible rails, steady as trains. This is what robots are good at: precision, predictability, consistency. That’s why they can drive so much better than imprecise, unpredictable, inconsistent humans.
“This technology can prevent drunk driving. Distracted driving. Drowsy driving,” says Medford, who spent his career in government advocating auto safety technology. “And there’s the issue of accessibility, too.”
Medford is a polished, reserved guy with a great poker face, a survivor of multiple White House regimes, but this is one time during our chat when he shows his emotional investment in the Google driverless project. He recounts the story of his neighbor in Bethesda, a woman in her eighties, long divorced, her children living in other cities. She depended on herself and did so magnificently, leading a full, active, and independent life. Her car was her freedom. An avid bridge player, she drove her less-independent friends over to play cards. She picked them up to go shopping. She drove them to attend events, making sure they, too, could get out and stay active. Then one day she was in a minor fender bender at the library. The police came and she received a ticket. A week later, she received a notice from the Department of Motor Vehicles that she would have to take a full driving test, including parallel parking, to retain her license. The worried woman begged Medford for help: she hadn’t parallel parked for years, she confessed. So he did what he had done for his teenage daughter years before, taking his sweet neighbor to a parking lot, setting up cones, and watched her try, and try, and try to do it. But she no longer had the range of motion in her neck to look back over her shoulder and steer and park at the same time. It was torture for her. Finally she got out of the car, weeping, and said, “I can’t do it anymore, Ron.” And she handed Medford the keys.
“That was the end of her independence,” he recounts quietly. “She stopped playing bridge. She stopped going out. She tried to use public transit, but she had a fall stepping down from a bus, and that was it. She ended up in a nursing home. It broke our hearts.”
He says he’s certain that autonomous car technology could have altered the course of his neighbor’s life, and that the potential for helping the elderly, the blind, and the physically challenged with mobility and independence is enormous. “This could be so meaningful for a large portion of the senior population. That’s a big part of why we’re pursuing this.”
It’s also why Google is more interested in moving faster than the car industry, Medford says.
That’s not the whole story, though. There is more than simple disagreement over design philosophies and the pace of transition to full automation at work here. This divide in approach is also about preserving a business model based on selling the maximum number of cars possible. Carmakers are investing in and researching autonomy in part as a defensive measure, because they know that, in some form, autonomous technology is inevitable. Someday it will be ubiquitous and, eventually, required. But if it is presented as a feature incorporated into otherwise traditional cars, then the current model of private car ownership, of two or three vehicles per average household in America, might continue. A slick autonomous driving option then becomes just another selling tactic, a point of competition that can enhance car sales. And by not requiring technology that makes a car driverless all the time, but just under certain conditions—on freeways with well-marked lanes, for instance—carmakers can take a less technically challenging route.
Google, which has no existing car business to protect, but sees endless potential for marketing its autonomous technology, has a very different vision. A car in which autonomy is not a feature but the essence of the vehicle, so that it can’t be driven by humans whether they want to or not, changes everything. Yes, it brings to an end the carnage of car crashes, of 35,000 deaths and 1.5 million trips to the ER every year. Yes, it gives mobility and independence to the elderly and the infirm who might otherwise be housebound. Yes, it reduces drunk driving and distracted driving to historical curiosities. But there’s so much more. There is the flip side of the fully autonomous car, the true differentiator between the carmakers’ evolutionary approach and Google’s attempt at revolution: the Google car can do stuff without any humans in the car at all.
The car that travels on its own can remedy each and every major problem facing the transportation system of systems and, along the way, end car ownership as we know it. That’s why transportation scholar, author, and blogger David Levinson of the University of Minnesota has proclaimed that “autonomous vehicles appear to be the next profound transportation technology.”1 That is, autonomous cars could be to regular cars what those same regular cars were to horses more than a century ago. That’s what Levinson foresees, and he is not alone.
Imagine this scenario: fully autonomous cars have become ubiquitous. This is doable sometime between 2030 and 2040 (with early adopters appearing by 2020). Human drivers will become little more than hobbyists, with car driving relegated to the same status as horseback riding: as recreation, not transportation. What does this transition do for the world? What might it look like?
Let’s say you open an app on your smartphone and summon a driverless car to your house. You need to get downtown to attend an all-day conference that starts in two hours. The app consults the latest crowdsourced traffic data and informs you the car will pick you up forty-five minutes before the conference starts to ensure an on-time arrival. At the appointed time, your phone buzzes: the car is outside your house. It takes you to the conference site, its route selected based on current traffic data; it drops you at the curb and then takes off to pick up another passenger. During the ride you read and answered e-mail, browsed the news, made a couple of phone calls, played some Words With Friends, and booked a dinner reservation. Drive time has become productive time. Neither you nor the autonomous car has to worry about parking at the end of the trip—a process that, in the once congested downtown area, used to be both time-consuming and expensive. Space once set aside for parking cars is now used more productively, in the forms of protected bike lanes, outdoor cafés, open space, and mini-parks. At the end of your conference, you use your app to schedule a pickup to return you home, your transportation needs met for the day—no fuss, no muss.
In this scenario, traditional car ownership makes no sense. If that was your personal car taking you downtown, then you would have to worry about finding and paying for parking, the driverless car cruising for a spot just like a human. And the same inefficiency that afflicts regular cars would kick in: the car would sit idle and unproductive all day. Instead, in a driverless rideshare scenario, any one of a buffet of options are available to you in seeking driverless car service: you could subscribe to a car plan much like a phone plan, buying minutes of travel rather than minutes of voice calls. Or perhaps payment is calculated by the mile. Or maybe you subscribe through a monthly fee like a data plan, with tiers of miles allowed, choosing a plan to match your needs. A thousand miles a month? Fifteen hundred? More? Or you might purchase any number of possible subscription plans, time-share plans, or contracts for autonomous vehicles on demand, with rates and terms kept reasonable by competition between providers. The key point here is that the car would belong to someone else, which might be a rideshare service, a car rental company, or the carmakers themselves, with customers paying only for what they need and use. Venues like Disneyland and Dodger Stadium and the Las Vegas strip might bundle tickets and hotel rooms with driverless service. Savvy local transit companies might get into the game, using driverless cars to solve the last-mile problem and get more people onto LA Metro’s light rail or the LA–San Francisco bullet train now under construction. Would you use mass transit more if a driverless car whisked you to the station just in time to board—and it saved you money as well as time? New York City’s transit company or Boston’s or Portland’s might enter such a market, simultaneously finding riders and easing traffic in their communities. They could offer multi-modal mileage plans, with savings for riders who take the bulk of their journeys on high-speed mass transit. Indeed, in this future scenario, carpool lanes could be converted to automated bus lanes; multiple driverless buses could use unerring robotic guidance systems to cruise bumper to bumper like trains at 150 miles an hour, drafting behind one another like race car drivers to cut wind drag and delivering passengers in record time.
For consumers, there would be no insurance costs for driverless cars in this scenario, no fuel costs, and no parking costs. What it would most resemble is a ridesharing service without the human behind the wheel, which would greatly lower the cost. And now you know why Uber, the global rideshare leader, which went from zero to ubiquitous since 2009, has hired away university researchers by the dozen and launched its own driverless car project. Robot Uber is on the drawing board. The company sees such a transformation of car culture as the inevitable outcome of a perfected driverless car.
This is why carmakers want autonomy to be evolutionary, not revolutionary, to keep that Uber robot at bay as long as possible. A world of fully autonomous cars à la Google or Uber would end car ownership as we know it.
It also ends cars as we know them. Think about it. The most common method of commuting to work in America—76 percent of the work trips we take —consists of one person in a car that could seat five or six people, that has an enormous trunk and a fuel tank with a range of hundreds of miles, all for a trip that, for the average American, is under fourteen miles each way.2 Indeed, the average car in American cities travels a total of 36.5 miles a day for all purposes, while the average rural-based car clocks in at 48.6 miles all day. As car owners, we want vehicles that can carry a lot of stuff and people long distances, even though we usually don’t do so, just so we’re covered when we need that capacity. Nobody wants too little car to do what they need to do. But that’s another built-in inefficiency of the current car-ownership economy. Not only are our cars parked most of the time, but they offer much more capacity than we need most of the time.
In the app-driven, on-demand scenario of driverless cars, the fleets for rent can be made up of a variety of vehicles purpose-built for each type of trip. Providers would build fleets out of a variety of vehicle types and sizes. Freed from the need to build cars that serve as general-purpose-vehicle versions of the Swiss Army knife, car designers could unleash a torrent of new designs for specific purposes: small one- or two-seat minis for short city trips; sleek in-line models for families of four seated one behind another for quick journeys to the beach; car bodies designed to easily store bicycles or surfboards or that have cargo space for trips to Costco. Robot shared vehicles could also usher in the rise of electric cars because range would no longer be an issue for most trips. Each car would serve multiple users in a day—instead of being parked twenty-two hours out of the day—on short trips with ample opportunities to recharge as needed between customers. The hundred-mile ranges of today’s electrics could be easily outstripped with lighter vehicles. Larger autonomous cars capable of carrying more people and stuff or going longer distances—hybrids or fuel-cell powered, perhaps—would be available as needed. Gasoline-powered cars in the consumer space would be dethroned.
In this scenario, fewer cars would be needed nationwide. Far fewer. And the drivers that would exist would flow in harmony instead of battling to pass and cut off and wave fists at one another. Congestion and traffic jams would be a bad memory; no more new lanes would have to be constructed, and those we have wouldn’t need to be twelve feet wide to accommodate human error. All those parking lots could become green spaces. We’d end up with smaller freeways, smaller streets, and much smaller transportation budgets for cities, states, and the federal government. These dollars could be redirected to maintaining and repairing what we have.
Some existing lanes could be given over to the goods-movement industry, which can also flourish with automation. Driver fatigue and rest rules would end. Safety would be guaranteed. Trucks could platoon—draft bumper to bumper—at high speeds in their dedicated lanes, cutting time and fuel. The robot Uber model may not replace ownership for freight and commercial vehicles, but all the other benefits of automation apply. For that matter, private ownership of robotic consumer cars would still continue for those who want it—and, at the outset, polls suggest many if not most Americans will not want to give up on a hundred years of pride, status, and habit linking them to car ownership. But the economics of a shared system and the practical benefits of having access to many different purpose-built cars instead of one jack-of-all-trades vehicle will be compelling. The Millennials will flock to it.
The upside for carmakers who seize this transformation as an opportunity instead of a threat are compelling, too: cars used around the clock instead of parked around the clock will wear out quickly and need to be replaced much more often than the ten-plus years Americans keep their privately owned cars today. Manufacturers of the new generation of purpose-built vehicles will replace the same car three or four times during those ten years. The first movers in this new competitive landscape will do very well by embracing this scenario. Today’s carmakers may end up making that leap. Or newcomers will swoop in and steal the business from them, as computer makers replaced the typewriter manufacturers.
Driverless cars are not a total panacea. Enormous amounts of transportation would still be embedded in our daily lives and consumer economy. Much of the current pain would be vanquished, though.
The problem of traffic jams would disappear.
The problem of traffic carnage would be gone, deaths dropping from the tens of thousands range to something measured in the hundreds or less. No more drunks on the road. No more texting and wandering over the centerline. No more careening into telephone poles on dark rural highways.
The problems of capacity, of overload on freeways, of trucks backed up at the ports, of cars crawling painfully up the 405 and every other overburdened highway at rush hour, all go away, too. As does the need to expend fantastic amounts of money expanding capacity.
For mass transit systems already in place, the last-mile problem can be solved at last, while compelling new forms of automated mass transit could emerge.
The pollution, climate, and health costs that have long been attached to our cars—and that drivers have never been forced to pay—go away, too. We get to keep our cars—a different ownership model, perhaps, yet still cars—but the transition from polluting fossil fuels to clean electric vehicles finally makes practical and economic sense in the land of robot cars. Fossil fuel dependence, with all its environmental, economic, and national security implications, could be eased.
The death of parking would be particularly huge. Parking isn’t just a drag on our time. It is a drag on the economy, a voracious land hog that grows with cancer-like relentlessness because most new real estate development includes a large, expensive, legally mandated parking component. Providing ample parking is a sensible-seeming policy but it comes with unintended consequences. For one, new parking in cities induces demand just like new freeway lanes, encouraging more people to drive into congested areas instead of making travel choices that reduce the need for parking (such as buses, subways, walking, or biking). Later, when tastes and trends inevitably shift and a once-thriving urban shopping center or other destination loses patrons, its under-utilized or abandoned parking lot will remain in place for years, an oil-stained island of asphalt and wasted space all too familiar to American city dwellers. A survey of mixed-used districts in cities across the country found that, despite a perception of parking scarcity at individual venues, the number of spaces available in the immediate areas exceeds demand by an average of 65 percent.3 Our oversupplied parking systems are yet another wildly inefficient use of resources in the door-to-door universe.
America’s parking king, Los Angeles County, is littered with disused lots, which is no surprise in an area where a fantastic 14 percent of the incorporated land area consists of parking. That’s the entire sprawling county of 87 suburbs ringing the City of Los Angeles, not just the concentrated urban core (where the percentage of land devoted to parking is closer to 30 percent). LA County boasts 18.6 million parking spaces in all. Nearly half that total consists of lots for business, industry, and government; 5.5 million spaces are taken up by off-street residential parking; and 3.6 million spaces can be found curbside.4 That adds up to 3.3 spaces for every car registered in the county. The space devoted to cars no one is driving covers about 200 square miles, which is 1.4 times the amount of land set aside for LA County freeways and streets. If those spaces were put together into a single parking lot, it would cover an area large enough to hold four San Franciscos.
The age of driverless cars would return all that land to community use. Los Angeles would have four San Franciscos of space to play with. Such an amount of land would be—is—priceless.
The rise of the robots would finally solve the problem of our bankrupt highway trust fund—not through policy brilliance but out of sheer starvation. The national gas tax that already falls so far short of paying drivers’ way would wither and die completely in a robot regime if, as many transportation scholars and researchers believe, the ascent of automation drives the rise of electric cars. No gasoline means no gas tax revenue at all, and Congress would be forced to find a new funding source. This would be the perfect opportunity to build user fees right into the emerging autonomous car sharing economy, just as airport funding is quietly built into every airline ticket today. It would be the drive more, pay more model we’re supposed to have right now, this time done right. And if politicians need cover for imposing such fees, all they have to say is this shift to automation will end the scourge of car violence, the number one killer of our children. How do you put a price tag on that?
The hardest part will be creating a fair and equitable road tax system during the decades of transition from the old world of human-driven gasoline-powered cars to the new autonomous model. The greatest benefits accrue only after most of the cars on the road are robotic. Most of the costs imposed on the roads—health, safety, and the environment—are caused by the old-school cars we’re driving now, which means, in a scrupulously fair world, their owners would pay a greater share of any new transportation tax than driverless electric car users during this transition. Polluters pay, that should be the rule. Safety risks pay. What could be more fair, more market-based than that? This would drive the shift to autonomy more quickly, but it would cause outrage, opposition, and real pain. Doing this part right and fairly would end up being much harder than building the technology itself.
Not everyone is entranced by this driverless car vision, to say the least. David A. Mindell, professor of engineering and the history of technology at the Massachusetts Institute of Technology, argues that the idea of full autonomy is impractical, even mythic, and that we would be better off striving for a perfect 50-50 blend of human and computer behind the wheel. He uses the 1969 Apollo moon landing as a prime example of the successful merging of human judgment and computer capabilities that led to a safe landing on the surface of the moon. Getting such a blending to work well and safely in everyday driving is a great technical challenge, Mindell argues, but the one most worth pursuing, because it keeps human judgment in the loop for those occasions when that enemy of robotic intelligence—the unexpected—arises.5
This is both a science-based argument and a seductive appeal to the human desire to retain control. But it doesn’t jibe with practical experience to date, which has shown that the most hazardous moment for autonomous cars is the hand-off from robot to human control, particularly during some sort of emergency. Startle reflexes, the split-second delay that causes humans to “freeze” before shifting from inattention to concentration to reaction, could make a partially automated system the worst, not the best, of both worlds. “The hardest part is this transition when we have partial automation,” Don Norman, director of the Design Lab at the University of California, San Diego, says. “This has been shown over and over and over again in aviation. People cannot keep their attention on what the task is. Because that is not the way we are built.”
A second problem is Mindell’s choice of examples: comparing the primitive Apollo moon-shot digital systems operated by the best trained space pilots in history to today’s technology with an ordinary driver at the wheel. That is a bit like comparing that same Apollo command module to the Wright Brothers biplane. It’s a mostly useless comparison, and not just because today’s computers and software are light-years ahead of Apollo’s or because NASA pilots are trained to pay attention far more rigorously than ordinary car drivers or even because they were concentrating on the mind-blowing moment of landing on the moon for the first time versus a somewhat less historic drive to the local Denny’s. No, the main problem with the comparison is that it ignores the critical role that detailed 3-D mapping of the driving environment plays in the success of the most advanced driverless car to date, the Google car. The moon landing was a first-time event. The Google car succeeds because it has “learned”—or been programmed—to safely navigate the most well-mapped terrain in the human universe, with over a million test miles under its digital belt, allowing it to adjust to the many variables that are added to that map by the presence of human drivers, cyclists, and pedestrians.
Even imperfect autonomy of this type will be far better than the distracted and imperfect human drivers roaming the highways today, Norman argues. “I am in favor of full automation. It’s inevitable and when it comes will save huge numbers of lives, and it will turn out to be a blessing.”
The Google car has mastered city driving. Grinding to a halt for a distracted computer-using jaywalker is flashy, but detecting and protecting a wayward pedestrian is actually a fairly easy challenge for the robot. Harder still is detecting a ball bouncing into the street, identifying it as a ball, then realizing that a child might follow that ball into the street and acting accordingly. The Google car can do that, too. But the most impressive achievement, wherein lies the true mastery of city driving, is how it handles the mundane but constantly varying minutiae of navigating through human traffic.
As I’m riding through the streets of Mountain View, just another car in the mix except for the cameras and radars and other distinctive moon-rover gear bolted to the Lexus test vehicle, the way the car handles a left turn blows me away. It approaches an intersection and signals a turn, then creeps forward without committing to the turn, just as a human does—not to get a better view, because this car sees everything, but to signal the intention to turn to other drivers. Then, the way clear, it executes the turn smoothly and moves on.
Exept for the fact that the driver is always attentive and never lead-footed, the experience is indistinguishable from driving with a human behind the wheel. And the reason for that is, in a way, a human is driving. The car doesn’t have a supercomputer or advanced artificial intelligence tucked inside. What it does have are lines and lines of code through which human beings have instructed the car what to do at an intersection, when a jaywalker appears, when construction workers block a lane of traffic. That’s why Google has driven its fleet of twenty-seven robotic test cars over a million highway and street miles throughout the Bay Area, and why they next moved on to Texas: to experience every possible situation and crisis on streets, so the car can be told by humans how to act. Google has even created what program director Chris Urmson calls his “red team”—a group of mischief makers who try to stump the car with unexpected obstacles or crazy behavior by other cars or cyclists who keep swerving out of their bike lanes. The robot just stops when it doesn’t know what to do. But each new line of code eliminates one more stopper. This is how the robot “learns”; it’s not about the hardware, which is fairly mature technology. It’s all about the software, which is why a Silicon Valley company like Google can compete with Detroit carmakers, most of which have struggled with making software as good as their hardware.
There are still obstacles. Lidar is confused by reflections. The car doesn’t know how to handle snow yet. Heavy rain is a problem, too. The spray of water that kicks up behind other vehicles creates a ghost image on the car’s sensors that can appear to be a solid object, something to brake for. Medford says the coders have to figure out how to filter those false positives. There is, he admits, a long to-do list. First among them is the creation of detailed three-dimensional maps that the car uses to navigate. This is the other secret sauce of the Google car: its mapping prowess. These are the Oxford English Dictionary of maps, far more detailed than the Google Maps on smartphones and home computers. They measure curb height and sidewalk width, they note bike lanes and median dividers, and they take special note of distinctive markers in the landscape to aid in navigation with far greater precision than GPS coordinates. The maps are not hard to make, Medford says: “You just drive the car in manual mode and the computer builds the map.” So far, the company has mapped all the freeways in the Bay Area and all the streets in the little city of Mountain View, population 77,846. All they have to do is map the rest of the country before bringing the car to market, Medford says. Most people might consider that daunting, but the folks at Google shrug it off as the least interesting challenge before them. “It’s just driving,” says Medford. “We’ll get to it after we get the technology perfected.”
There have been several accidents involving the Google car—none serious, and none, according to Google, the robot’s fault. This is how the Google car’s insistence on observing the speed limit and not running red lights can be a safety issue. It’s been rear-ended several times, once because it was observing the speed limit on a street full of speeders, and once when it was properly stopped at an intersection when a distracted human plowed into it. Humans in both cars complained of neck pain afterward, but there were no serious injuries. The mix of autonomous and human-driven cars on the road may reduce accidents, but the greatest gains in safety likely will not occur until the vast majority of vehicles are driverless, Medford says.
The Google car has been pulled over by the police only once: for driving too slowly. It was traveling 25 miles per hour in a 35 mile per hour zone on a four-lane street in Mountain View (which is not technically illegal). The car was not malfunctioning: it is programmed to never exceed 25 miles per hour. The incident went viral, with the little Mountain View police department flooded with inquiries from media around the world. The car got off with a warning.
There’s another wrinkle in this transition to vehicles smarter than the humans inside them: the idea of connected cars. This is a technology being pushed by the U.S. Department of Transportation, among others. It’s different than driverless systems, but complementary to them. The idea is to leverage wireless technology to connect one vehicle to another—the digerati call this V2V—and to connect cars to the road and street infrastructure—V2I. Connecting cars would be a relatively simple matter, using inexpensive and proven transponders similar to the ones built into every airplane for decades, which transmit position, direction, and speed. These beacons could be anonymous if privacy concerns impede adoption, but the key would be how they could make an autonomous car aware of other vehicles that its sensors might not detect because of range, obstruction, or most relevantly, bad weather. This would be particularly valuable in a transitional mobility universe where autonomous cars and human drivers are mixing it up, as the robot cars would always know in advance if a human driver was slowing down to stop at an intersection—or on a course to just blow through.
The V2I concept would not be so easy to implement, because it would take many years and dollars to wire the built landscape and build beacons into our traffic signals, buildings, parking structures, and road signs. But such tech would solve the problem of sensor blindness in rain, fog, and snow. The road could literally talk to the cars, and let the autonomous driver know what’s being obscured by adverse conditions.
There’s no timetable or money at present for getting V2V or V2I up and running, and so all the driverless car projects that are being prepared for market now are designed to operate without any connection to anything—they are intended to be self-sufficient. This has the added bonus of greater digital security, as networked cars, like computers linked to the Internet, would be a far easier target to hack. But these same driverless cars will be able to jack into the connected car and infrastructure systems someday (if such systems ever become available), so security will be a continuing concern.
However it progresses, the transition to driverless cars will not be without some bumps and bruises, if only metaphorically. New technology and change always bring pain with the gains. For all its faults and harm, the hundred-year reign of the traditional car has had a powerful and positive impact on the economy and society, bringing unprecedented mobility and shaping our human landscape, culture, work, and lives for generations. This is part of the miracle of our door-to-door system of systems, and it is almost impossible to imagine replacing it with something else. Cars are linked in mind and deed with Americans’ sense of personal freedom and opportunity.
But the current transportation regime is, literally, killing us—with the short-term bang and shatter of crashes, and the longer-term poison of pollution and carbon and oil dependence. Its inefficiencies cost us economically—more than we can pay for now, much less in the future.
The driverless future that could ease the killing and the costs, if it comes, will bring another kind of pain: lost jobs. Robots behind the wheels of cars and trucks will eliminate millions of human jobs and billions in salaries. Disruptive technologies always create new winners and losers, but it’s not clear how or if the advent of automation will create enough new jobs to take the place of those lost. This is not new. Refrigerators put the ice man and his horse-drawn carriage out of business less than a century ago. Automatic washers put the laundry business down. Kodak’s entire business used to be based on people shipping their cameras to the company to have the film inside developed and reloaded, then later customers just sent in the film. Now film photography is a near-dead business. Digital photography took a thriving analog technology with transportation deeply embedded within it and replaced it with a model that requires no transportation at all once the product is purchased. And now phones have eviscerated the stand-alone digital camera business as well.
The replacement of truck, bus, and cab drivers with automation will be wrenching, particularly since taxis have become an entry point into the workforce for immigrants, and truck and bus driving have provided one of the few enduring and plentiful blue-collar jobs that still provide reliable paths to middle-class prosperity. The American Trucking Associations reports that there are about 3 million truck drivers working in the U.S.—it’s the single most common job in a majority of states—and about 1.7 million of that number are long-haul truckers, who would be most vulnerable to displacement by autonomous technology. The number of long-haul truckers is expected to grow by a minimum of 11 percent by 2022.6 But the economic case for trucking fleets to go robotic is too powerful to ignore; this will happen, and likely far faster than consumer autos convert. This will be hard, but no more so than the fate of the once-robust web of smiths, farriers, horse dealers, feed stores, veterinarians, trainers, breeders, and stables that serviced every town and city in America until the 1920s, when human-driven cars displaced horses as the king of personal and commercial transport. Prior to that, during “peak horse,” there was one working horse for every three Americans, and in New York City there were 297 horse-cart trips a year per person.7 This transition was so recent that it occurred as the parents of the baby boom generation were growing up. Hundreds of thousands of jobs and thousands of businesses were wiped out when engines replaced horses as the backbone of human, freight, and farm transportation. This is what technological breakthroughs do. That’s why they call it disruption. It’s hard. It’s upsetting.
But so are 35,000 deaths a year. So are 2.5 million trips to the ER a year. So are 5 million collisions a year. The single greatest cause of death for Americans ages one to thirty—that’s disruption, too. That’s pain. How much is ending that pain worth, particularly when the same solution also can take our roads and bridges out of overload and bankruptcy, toxins out of our air, and greenhouse gases out of the climate?
Still, Americans are uneasy about such change, and polls show most trust human drivers more than machines.8 They may balk at such a transformation because cars, so long a part of the landscape and language of daily life, have become more than just transportation. We may hate the travails of transportation. We may hate the time we are stranded in traffic inside our cars. But we love our cars, that ultimate shipping container in which we ship ourselves and our families. We love them not as transportation tools but as objects and comforts and statements of style, the way a carpenter loves a well-made, well-used hand tool and is heartbroken when it’s lost. It is the alchemy of cars we love, the interaction between hand and wheel and road, the gliding pleasure of a perfectly negotiated banked curve, the smooth and timeless sensation of having a world to yourself while driving at night, a sweet sound system swathing you in music and the solitude of movement. And, yes, for some—those who can afford it, or at least borrow to achieve it—there is the pleasure of having a trophy car that others may envy.
Many, perhaps most, American drivers will not want to give up these pleasures, these luxuries. But would they have to give them up in a world dominated by on-demand autonomous transportation appliances?
The short answer is no. First, any shift will be slow. The most optimistic predictions put 2030 as the soonest driverless cars are likely to become dominant. This would require firm action by government and industry to help make it happen. Given the inability of the government to add even a nickel to the gas tax to keep our bridges from collapsing, this seems unlikely. Trucking and delivery fleets may transition to autonomous driving that fast, because of the incredible economic benefits, and that may push passenger cars to make the leap more quickly. The year 2040 is a good bet for the rise of the on-demand robot passenger car. Whatever the time frame, with 265 million cars on the road in America, changing out that fleet would not and could not happen quickly or en masse. There will be a gradual transition.
Even then, just as people who find pleasure and fulfillment in riding horses haven’t had to give that up, drivers could continue to enjoy taking the wheel. The evolution of horse travel provides the road map for the future of the human-driven car, the difference between transportation and recreation. Horses were once beasts of burden. They were our motors. Oats were big business, the horse era’s equivalent of gasoline today (except the oats were a form of renewable energy). Now horseback riding is a luxury, enjoyed not on the regular streets and highways they gave up to cars and trucks, but in parks, on trails, and in equine-friendly communities. Cars will follow the same pattern. If there is a demand for it, there will be car parks and preserves and safe roads where humans can drive manually away from regular traffic. And if users want luxurious robot cars, the fleet owners will fill that demand, too. It will cost more, just as buying a regular luxury car costs more, but that’s a model we’re already used to. Far from denying drivers the joys they now find, the driverless scenario of the future could be the best of both worlds—cleaner, more efficient, less costly, and far less deadly.
It may be that automated vehicles’ utility and opportunity will not be fully understood until they are out there in force. The smartphone experience may be instructive here. When Apple introduced the first iPhone, the design did not include open access to independent app developers. It was only when later iterations opened up the platform to outside innovators who thought up novel and new uses for smartphones that the true potential of such devices became clear. Autonomous cars are first and foremost a software product.
Google has designed with the car component maker Bosch and other partners a two-seater electric bubble of a car with no steering wheel or pedals, purpose built for short-distance city driving with a maximum speed of twenty-five miles an hour from its battery electric propulsion. It’s cute and nonthreatening; the idea is to choose a city or two to test-market a robot car capable of transporting people in urban areas, and then go from there. Meanwhile, the big carmakers are focusing primarily on the highway automation feature; in the next few years, the future of door-to-door will be on display, and the next big change—as big as the first car or the first steam locomotive or the first airplane—will begin.
Google is a bit cagey about its own time frame. But it’s not far off, says project leader Urmson, one of the small cadre of visionary engineers who have been driving robot car development since the first Defense Department competition in 2005 sent college teams into the desert with the first experiments in autonomous driving. Urmson has a simple goal: he says he doesn’t want his son to have a driver’s license.
His son turned eleven in 2015.