MY TAXI RIDE TAKES ME from the central shopping district of Singapore to a university campus — and into the future. Passing dense clusters of high-rise buildings, the driver weaves between lanes on a smooth new expressway equipped with electronic sensors that administer surcharges for driving during peak times. Below ground is an advanced and extensive train network. Sitting beside me in the back of the cab is Emilio Frazzoli, an Italian-born aeronautics specialist and MIT professor who is helping Singapore’s government prepare for a new dawn in transportation. “They complain about traffic congestion here. But they don’t know congestion,” he says, contrasting the briskly moving highway with his rush-hour commute in Cambridge, when it can take 40 minutes to travel 4 miles.
The government’s intolerance of traffic tie-ups helps explain why Singapore is today better positioned than most countries along the road to tomorrow. Soon enough, Frazzoli tells me, taxis like ours will become ride-sharing vehicles, powered by electricity rather than gas. Also: There will be no one behind the steering wheel. And no steering wheel.
Singapore, the tiny, humming city-state sandwiched between Malaysia and Indonesia that’s known for its sleek skyscrapers, paternalistic government, and curious war against chewing gum, is at once gloriously forward-looking and more than a smidge dystopian. It’s also the perfect place to advance the technology for driverless cars.
Frazzoli’s team of researchers from the Massachusetts Institute of Technology and the National University of Singapore has been testing autonomous vehicles on a limited number of campus roads. But in a few months, the government is expected to allow the testing to expand to Singapore’s sprawling “One-North” district, a hub for biotech, media, and R&D companies. One section of this architecturally vibrant if antiseptic neighborhood is called, in a name right out of Epcot Center, Fusionopolis. It features the horseshoe-shaped, glass-encased Asian headquarters of animation and visual effects powerhouse Lucasfilm, and the building draws its design and name, The Sandcrawler, from the mobile fortress featured in Star Wars. Yet when Frazzoli’s fleet begins circling the futuristic buildings of this futuristic neighborhood, the autonomous vehicles will encounter present-day conditions, interacting with conventional cars and regular drivers on public roads.
This demonstration project comes during a year of tremendous progress — and even greater hype — for self-driving vehicles, a world that has finally launched out of the realm of beard-stroking futurists and dreamy science-fiction writers. These days, barely a week goes by without news about some company taking us another step closer to a driver-free future. Tesla, Uber, Mercedes, Audi, Daimler, Delphi, Honda, Nissan, GM, Volvo, Apple, and Google, especially Google.
Things are moving so fast that there’s no consensus yet on whether we really want our cars to drive themselves — or even what we should call them. Should the term be driverless? Self-driving? Autonomous? Automated? Robo? This naming confusion betrays just how early we still are in our journey, despite all the triumphant headlines. One tech thinker compares the present day with the “horseless carriage” period in the early 20th century, when the transformative power of the automobile remained obscured because we could see the innovation only in relation to the technology it was about to displace.
How long it takes us to get through this period will largely be determined by how well the field’s leaders are able to clear the remaining hurdles of technology, regulation and liability, and societal acceptance. Will engineers be able to make self-driving cars intuitive enough to avoid introducing new dangers — and lawsuits? Will people willingly hand over their keys and trust their lives to a robot?
To try to get some answers on how things might actually shake out, I traveled to two distant spots where researchers are using very different approaches. As I found both in Singapore and at Google headquarters in California, the future is probably closer than most of us know. It’s also probably still further from widespread acceptance than many tech evangelists would have us believe.
THE LOOKS I ATTRACT from bystanders are priceless, a blend of curiosity, agitation, and distrust — the same mix of emotions I feel when approaching any new film by Wes Anderson. Some people jump back, startled. Some smile on, befuddled. Others simply point and stare.
I’m sitting in the cream-colored passenger’s seat of a green electric golf cart as it tools around the modern Campus for Research Excellence and Technological Enterprise, or CREATE, a hub for researchers that Singapore has attracted through partnerships with MIT and other top universities in half a dozen countries.
It’s not that people on campus have never seen a golf cart before. It’s that the vehicle is powering itself to its destination with evident purpose but without a soul in the driver’s seat. In place of a steering wheel, there is a monitor displaying a dynamic map of the area. It’s synched to the map on the screen of the smartphone in my hand. I had used the phone app’s drop-down menu to arrange the pickup and drop-off location for my ride.
There’s a flashing yellow light on the roof of the cart, a spinning laser on the front bumper, and a pair of sensors attached to the rear side panels. The vehicle glides along to John Williams’s swelling soundtrack from Star Wars. At first, it feels unsettling to sit in the passenger’s seat of a moving driverless vehicle. The first time it approaches an unsuspecting pedestrian with her face buried in her phone, I’m tempted to yell “Look out!” before it plows her over. But maybe because the vehicle isn’t moving very fast, or maybe because I see Frazzoli standing nearby with an easy smile on his face, I keep my mouth shut. The golf cart pauses, the pedestrian looks up and freezes, and then we both watch with relief as it swerves away from her.
Like a NASCAR entrant, the green cart is emblazoned with a host of logos. Rather than Coors Light and Sunoco, though, these sponsors are all academic: MIT, the National University of Singapore, and the Singapore-MIT Alliance for Research and Technology, or SMART.
The vehicle has chrome-colored pedals for acceleration and braking, but the brake pedal operates itself like the foot pedals on a player piano. (There is a red emergency kill switch in the center of the dashboard, just in case.)
As with many working on self-driving vehicles today, Frazzoli was inspired through his involvement in the 2007 DARPA Urban Challenge, run by the R&D arm of the US Defense Department. The challenge’s goal was to take the autonomous technology designed to resupply military convoys in desert battlefields and adapt it to the civilian world of side streets and stoplights. That contest became a Woodstock for roboticists and other engineers.
Frazzoli was so encouraged by the results that he shifted his research focus from pilotless drones to driverless cars. “I realized we were sitting on something much bigger than I had imagined,” he says, “something that had the potential to change everybody’s life.”
To avoid the trap of prohibitively expensive technology, Frazzoli and his team set about trying to get the job done using relatively cheap, off-the-shelf components.
Others in the field began working on either entirely driverless consumer vehicles or vehicles that would rely on human drivers only in a pinch. Frazzoli, now 45, had his doubts about both tracks. He knew from his aeronautics background that safety lapses had increased after autopilot technology was introduced in airplanes. It’s hard to get humans to remain alert for a possible crisis situation if most of the time they know they are not needed. Besides, requiring drivers to remain on alert for the entire trip would deny them the ability to get other work done, whether that means e-mailing a client, reading to a child, or taking a nap.
Getting rid of the driver altogether could produce enormous leaps in productivity, accessibility, and safety. After all, these vehicles would allow people to work while in transit. They would make cars available to the disabled, the elderly, and the inebriated. And they would go a long way toward eliminating the cause of more than 90 percent of car accidents today: human error.
Yet Frazzoli figured it could take a really long time to get the technology reliable and affordable enough to make self-driving cars available on a mass scale. He concluded that it would be easier, faster, and much better for the environment to focus on creating ride-sharing fleets of electric, self-driving vehicles — a kind of autonomous Zipcar. You would take one home from the train station and it would drive itself back for the next commuter. These fleets could solve the “first mile, last mile” problem that bedevils many dense cities. Even places like Singapore with highly reliable train networks find that many commuters opt to drive simply because there are no efficient and affordable options to get them from home to the train station, or from the train station to their workplace. This increases congestion, pollution, and wasted real estate in the form of parking garages where these cars sit idle for most of the day.
Although Frazzoli’s team has far fewer resources than many corporate players in the self-driving sphere, it has one enormous advantage: Singapore as its test market. Here is a tidy, thoroughly modern nation with a sophisticated transit infrastructure (including a few self-driving trains). Moreover, it has a highly centralized government that has already conditioned its population of 5.5 million to accept driving restrictions that would be unthinkable in the United States.
Many Singaporeans tend to behave less like citizens of a freewheeling democracy and more like employees of a benevolent, publicly minded corporation. In the same way the government has aggressively used economics to keep the country clean, with steep fines for not flushing public toilets or for selling gum, it has kept a lid on traffic through congestion surcharges and by making it extremely expensive for people to own cars. Residents must pay $60,000 or more just for permission to own a car, and that certificate lasts for only 10 years. Then there are crazy import taxes. A basic Toyota Prius can cost a Singaporean more than $150,000.
These restrictions would seem onerous to Americans, but they represent ideal levers for the introduction of fleets of shared autonomous vehicles. The biggest challenge for bringing self-driving cars into the mainstream is that these vehicles will need to share the road with unpredictable and often illogical human drivers. But, Frazzoli says, Singapore could one day simply designate entire stretches of its expressways for self-driving vehicles.
Last fall, Frazzoli’s team ran a six-day trial with the self-driving carts at Singapore’s Chinese and Japanese Gardens. Using a Web system to book trips to one of the 10 stations scattered around the gardens, visitors took more than 200 rides. A member of the lab team rode on a bike alongside the cart for each ride, in case of emergency, but there were no bad outcomes besides one smushed frog.
Instead of GPS, the vehicles use more finely detailed localization technology relying on cameras and LIDAR (light detection and ranging) sensors to pinpoint where the vehicle is and determine what’s around it. Vehicles are first driven manually along the terrain to build the detailed maps needed for autonomous driving. Engineers in the lab work to refine the algorithms so the computer can identify all the players in the car’s path and try to anticipate their actions.
Frazzoli’s team uses pretty much the same technology to operate its one autonomous car — a compact electric Mitsubishi i-MiEV. Just before my visit, one of its off-the-shelf parts malfunctioned, which meant my test-drive in the car would have to be manual. Still, as I venture out onto the campus roads, my nearly 6-foot-4 frame crammed into the tiny driver’s area, it’s not hard to imagine a fleet of these cars tooling around Singapore before long.
Then again, there are always more stops than you expect on the road to the future. Despite the strong endorsement that the project has received from Singapore’s top transportation officials, there’s still a bureaucracy to navigate. I sit in on a meeting between Frazzoli’s team and some mid-level bureaucrats from the country’s equivalent of the Registry of Motor Vehicles. Frazzoli goes into the meeting hoping it might end with an official start date for the rollout of self-driving vehicles in the One-North neighborhood.
James Fu, a native Singaporean who serves as Frazzoli’s project leader, begins taking the group through a 10-slide presentation summarizing the project’s lofty ambitions and cutting-edge technology. Slide number four features a photo of Michael J. Fox in Back to the Future II. Fans of that 1989 film will recall that Marty McFly’s DeLorean transported him to a time that seemed impossibly distant back then: the year 2015. But what could have been a quick meeting drags out for more than two hours, and it ends without a resolution. Fu never gets past slide number four, which is labeled “The Future is NOW.”
ON MY WAY HOME FROM SINGAPORE I stop in Mountain View, California. During my last visit to Google a dozen years ago, the company had three buildings and one cafeteria. Today, it sprawls over block after block of town. It also has 35 cafeterias — each serving such a tempting menu of free organic food that new employees complain about putting on the “Google 15.”
On the day of my visit, employees of Google X — the company’s experimental R&D unit that houses the team for the self-driving car — are packing up for a move to a former shopping mall that Google has taken over. The similarities between life at Google and life in Singapore are impossible to ignore. In both of these shiny places, the people are given abundant resources and encouraged to work hard to advance themselves and the mother corporation, but they also seem acutely aware of the need to remain on message and not run afoul of management.
There are two trained safety drivers with me on my test ride in the Google self-driving Lexus SUV. Behind the wheel is 29-year-old Greg Hanabusa. When people ask what he does for living, he says, “I watch a car drive itself.” In the passenger seat, with a laptop resting on her knees, is Alyssa Bowman, a 31-year-old former middle school social studies teacher. They’ve both spent the last few years in test vehicles touring the streets of Mountain View.
To disable the self-driving mechanism, all Hanabusa has to do is touch the steering wheel or tap on the brakes and a voice purrs, “Manual control.” Bowman’s job is to monitor the laptop, which codes various roadway objects as colored boxes: pedestrians appear in yellow, bicyclists in red, and vehicles in either green or pink. Sensors and cameras feed the car’s main computer while Google software identifies nearby objects by their shape and movement patterns and then determines how the vehicle will respond. The software has become sophisticated enough to discern even a bicyclist’s hand signals.
During my ride, the car is occasionally jerky, seeming to sense trouble even where it doesn’t exist. This is especially true in parking lots, which Bowman admits remain a challenge for the algorithm, given the slow speeds and unpredictable driving patterns there. But what’s remarkable is how generally smooth and uneventful the ride is. When a kid pedaling a bike well ahead of us swerves into our lane, the Google car slows down and drifts away from him. When there is a lane closure because of construction on a side street, the car reroutes us, even before the construction is visible to our eyes. The technology of tomorrow is here now operating daily on the streets of this Silicon Valley town.
As we turn right onto a side street, the Lexus — identifiable by its circling rooftop laser and prominent lettering that reads “Google” and “self-driving car” — has to veer over the center line to avoid hitting a car parked dangerously close to the intersection. The driver is standing next to the car, using his iPhone to capture on video the curiosity that is the Google car.
There is a mechanism on Bowman’s laptop that allows her to give instant feedback to the engineers if she notices a glitch. Because human drivers sometimes blow red lights, Google cars idling at stoplights — in response to feedback from test-drives — now automatically wait 1.5 seconds after the light turns green before proceeding through the intersection.
Later that same afternoon, when I meet with the head of Google’s self-driving team, Chris Urmson, I ask him about that safety upgrade. “Waiting a second and a half at green lights sounds great,” I say, “but doing that in Boston will get you rear-ended by the impatient driver behind you.” Since Google hopes to make these self-driving cars available nationwide, I ask: Will they be smart enough to adapt to the driving culture of particular regions of the country?
“That’s something we will have to work on,” says Urmson. A Canadian by birth, he studied engineering at Carnegie Mellon University, where he learned about an ingrained local display of courtesy called the “Pittsburgh left.” Eventually, he says, Google self-driving cars will need to know to adopt this courtesy if they’re in Pittsburgh but be more coldhearted if they’re in Boston.
Urmson is an upbeat 38-year-old and veteran of the DARPA Challenge who bikes to work most days. By about three years ago, his team had refined technology to take over for drivers on highways, once the driver had gotten the car into position. It was essentially a super-intuitive version of cruise control, somewhat like the autopilot technology that Tesla is in the process of rolling out. In both systems, the human driver would step in during a crisis but otherwise would leave matters to the robot.
Despite positive reviews for Google’s highway assist, Urmson’s team found that drivers became too trusting and, therefore, too inattentive. He then decided to shift gears and go for the “moonshot,” to eliminate the driver altogether.
They began testing retrofitted Toyota Priuses and Lexus SUV’s on the side streets of Mountain View, but eventually built a self-driving vehicle from the ground up. This summer, Google will begin testing these prototype vehicles, which look a bit like Volkswagen bugs, on public roads. They were designed with no steering wheel or brake pedal, though those devices will be added for this testing period.
Safety is a big selling point for Google. Ron Medford, the team’s safety director, points out that there are 5.6 million car accidents reported to police each year in the United States, including more than 33,000 fatalities. “About one-third of fatalities are alcohol-related. Many involve distracted drivers, sleepy drivers, and drivers using excessive speed,” he says. “Our cars won’t do any of that.”
It’s no surprise that Google hired Medford to make that safety case to government regulators, since the 66-year-old previously served as deputy administrator of the National Highway Traffic Safety Administration.
A throbbing question about self-driving cars is who will be held responsible in the case of an accident: the owner/passenger or the manufacturer? Urmson predicts that insurance companies will eventually shift from insuring individual drivers to writing product liability policies for car manufacturers. Rather than trying to predict the behavior of individual drivers, he says, insurers could make smarter bets using data about a specific model’s performance.
Urmson says the biggest remaining challenge will likely be persuading lots of people that this is a product they can trust and actually want. By now, he’s heard all the arguments about Americans’ long love affair with the open road. But he counters that being stuck on the freeway is fun for no one, not even someone driving a Porsche. Self-driving cars will eliminate all that tedium. And for parents fatigued from shuttling kids to one activity after another, his vision sounds particularly tempting: plop them in a self-driving car, plug in the coordinates, and they’ll be good to go until the car brings them home for supper.
While Urmson continues trying to expand people’s imaginations about how good the future could be, he knows it’s just as important for him to guard against unrealistic expectations.
Even if self-driving cars usher in much safer roadways, he knows there will still be accidents. If people expect perfection, though, a few high-profile fatalities could shake, if not shred, public confidence in this new technology. (There are legitimate concerns about vulnerability to hacking. And anyone who expects the Teamsters to just sit back and accept the arrival of the self-driving 18-wheeler freight trucks now being developed clearly hasn’t watched enough Scorsese films.)
That’s why when, a week after my visit, the Associated Press reported that there had been at least three accidents in Google test cars in recent years, Urmson wrote a lengthy blog post revealing that there had actually been 11 accidents over the last six years. (He did point out, however, that the Google vehicles had not been at fault.)
While there are any number of ways this could all play out, Urmson has no doubt about the direction we’re headed. “I’ve gone from hoping it will happen, to thinking it will happen, to knowing it’s going to happen.”
Do-it-themselves cars at a glance
AFTER TRAVELING MORE THAN 20,000 MILES and spending several days in flight, all to see the future, I need to get home — and back to reality. On a sunny morning in May, I am sitting shotgun in John Leonard’s 2004 Buick Century, as he tries to nudge the gray sedan out of a Newton side street to make a left turn onto busy Parker Street during rush hour.
Leonard is an MIT professor of mechanical engineering who’s been working in robotics for nearly three decades. He’s a 50-year-old who retains from his youth both his reddish hair and his enthusiasm about the promise of technology. That’s why he’s so uncomfortable with his current role, as splasher-of-cold-water on what he sees as the overly optimistic predictions now animating his field.
The flow of cars along Parker is so dense in both directions that there is no opening for Leonard to turn left onto it from the side street.
Leonard does what most of us around Boston have been conditioned to do in this situation, inching his way into the intersection. Several times, a charitable driver in the northbound lane leaves room for him to cross, but he can’t find a big enough opening in the southbound lane and the charity expires. Finally, Leonard hits the gas and the Buick jolts toward the center line, forcing the oncoming southbound driver to hit the brakes. “Sorry,” he says through his window.
It takes a good three minutes for him to make his turn. “My guess is a robotic car would just wait and wait for a long time,” he says.
We head toward Countryside Elementary School, which sits at a difficult intersection made more difficult by the morning-zoo nature of some parents dropping off their kids wherever it’s most convenient for them, rather than what’s best for the common good.
Leonard turns to me and says, “Can you imagine putting your elementary school kid into an autonomous car without an adult and into a complex situation like morning drop-off?” Suddenly, the prospect of this brave new liberated world feels a lot scarier than it did when I’d been pondering it with the cheery folks at Google.
Add to that the problem of sensors not working in snow or heavy rain and being unable to decipher obscured road markings, and the reasons for Leonard’s caution becomes clearer. On top of that, these self-driving vehicles would be asked to interpret highly variable hand signals from police and crossing guards and interact with cars driven by sometimes unhinged humans, like the driver I saw a day earlier backing up several hundred yards in the high-speed lane on Route 128 because he had missed an exit.
Leonard admires Chris Urmson’s talents and Google’s vast resources. When he rode in one of Google’s self-driving Lexuses last year, he says, “I felt like I was on the beach at Kitty Hawk.” But he points out that although the Wright brothers got airborne in 1903, it wasn’t until 1958 that regular trans-Atlantic jet service began.
While Urmson is betting that completely self-driving cars will be ready for the public in five years, Leonard says, “I think you need to add a zero to that figure.”
The oldest of Leonard’s three sons is in eighth grade. He’s hoping that by the time that boy is ready to drive, a less sexy but more realistic path to improved safety will be widely available, something he calls “hidden autonomy.” Rather than a robot controlling the car all of the time, Leonard describes a “system that will aggressively jump in and save you before an accident happens.” Several carmakers already offer emergency braking and similar automated features. But truly making driving safer, he says, will require much smarter technology.
I share with him the point that both Urmson and Frazzoli had made — about perfection being the enemy of good. Yes, Urmson admits, they haven’t figured out how to make the self-driving cars work in the snow or heavy rain. “But guess what? There are many parts of the country where it never snows. In California, we’re in the middle of a historic drought.”
In other words, unless we start with limited rollouts, in places where it makes sense, we may never get out of neutral. With Frazzoli’s vision for a fleet of ride-sharing vehicles, maybe ex-Uber drivers could be on call to operate the fleet manually during bad weather — for a surge-priced premium.
As the morning rush hour begins to subside, Leonard maneuvers his Buick back onto the side street where our tour began. After he pulls over to let me out, a black Mercedes sidles up next to us, and the passenger’s window lowers. The driver announces to Leonard, “You’ve got a flat.”
We both hop out of the car. Staring at that deflated rubber tube, I can’t help but wonder: Maybe it’s time to get some bright minds working on a self-changing tire.