Here in South Boston, sandwiched between a dry dock to the north and the port terminal to the south, a small, silent car slips quietly along the street. Exceedingly polite, it moves patiently between the more numerous and far more massive tractor-trailers and delivery trucks that populate the maritime park’s roads. Someone is in the driver’s seat, but appearances can be deceiving. The man behind the wheel is merely there to make sure everything is going as planned.
If it weren’t for the metal canister sprouting from the peak of its roof and the small, black hockey pucks jutting out from the front fenders, you’d probably think this was just another company car out on an errand.
Instead, it’s a relatively low-key introduction of three technologies that could fundamentally change the way we get around—autonomy, ride sharing, and electrification.
The car is one of two that began prowling this remote corner of Boston a few weeks ago, the small test fleet of nuTonomy, a Cambridge-based startup that’s developing an autonomous car-sharing system. The car’s autonomous capabilities are the most obvious of the three technologies and certainly the most captivating. They’re also the linchpin that could allow the other two—ride sharing and electrification—become part and parcel of mobility in the 21st century.
But that cascade—where autonomy drives the adoption of ride sharing which helps spread electrification—is by no means guaranteed. In fact, it’s the rapid development of autonomous vehicles that some transportation experts fear—not because of the safety concerns, though many experts acknowledge the perilous ’tween phase where the computer occasionally needs the help of a distractible human, nor because of their potential to cause mass unemployment, though there, too, they know the intractability of the problem.
Instead they’re worried that fully autonomous features will arrive too soon, that everyone will want one, and that they’ll be so cheap that nearly everyone will be able to afford them. For transportation experts, that’s a nightmare scenario. In fact, they call it “hell.”
Here’s why: If self-driving cars become commonplace without other changes to our cars or how we use them, the roads of the future could be hellish indeed. People will spend more time on the road because they can do stuff other than drive, clogging highways and arteries. Emissions will spike as people travel farther and sit longer in traffic jams. Such a future won’t be a leap forward for automotive transportation, but a distilled version of the worst of today’s car culture.
But, the experts say, hell isn’t our only possible destination. Self-driving cars could deliver us a comparatively utopian future, where trips are quicker and the environmental impact is far lower. The financial cost to drive will be so low that it could unleash a new wave of mobility and all the benefits that come with it. To get to this “heaven,” as experts call it, fully autonomous cars need to be widely adopted at almost exactly the same time as the other two technologies—ride sharing and electrification. If that happens, we’ll be able to reduce the number of cars on the road, thin out traffic jams, and, with clean sources of energy, slash emissions, saving thousands of lives every year and curtailing at least some of the risk of dangerous climate change.
“This intersection of autonomous, shared, and electric is what could potentially get us to what a lot of transportation and energy analysts see as a kind of carbon nirvana for transportation,” says Levi Tillemann, a managing partner at Valence Strategic and author of The Great Race: The Global Quest for the Car of the Future.
Getting there won’t be easy, though. To reach nirvana, we’ll need dramatic advances in sensors and artificial intelligence, breakthroughs in battery capacity and charging speeds, deeply sophisticated routing algorithms, and widespread social acceptance of sharing a ride. All in the next decade. It’ll be one hell of a ride.
Send In the Robots
Ten years ago, an ungainly looking Chevy Tahoe crept across the finish line at the 2007 DARPA Grand Challenge. The car bristled with millions of dollars worth of sensors and receivers, including radar and lidar sensors, high-resolution cameras, and GPS receivers. To integrate all that data and act on it, the truck was outfitted with what is essentially a small data center.
Since then, computers have grown more powerful, sensors cheaper and more accurate, and artificial intelligence more capable and independent. Despite those advances, full autonomy is still five to ten years off, according to the experts I spoke with. The biggest limitations are related to the artificial intelligences that power self-driving cars and the sensors that feed those AIs.
Most current autonomous vehicles rely on a type of AI known as deep learning, which uses mountains of data to train its algorithms. Self-driving cars teach themselves by observing similar situations over and over again. Take a pedestrian at a crosswalk. After digesting data on car-pedestrian interactions, an autonomous vehicle can reasonably guess how the next person it encounters will behave. But because no two pedestrians are alike, the algorithms must be exposed to dozens of interactions and be able to improvise to some extent. When nuTonomy introduces a new scenario to their software, says Karl Iagnemma, the company’s CEO, “we perform more than 100 simulated tests of that particular scenario, each under slightly different conditions.”
The same is true of just about everything else self-driving cars will encounter. Intersections differ from city to city, signage varies from country to country. Once you throw in weather conditions, the number of different situations a car will encounter is seemingly infinite.
“One of our cars drove 50 miles last week,” Iagnemma says. “It was just driving in circles, but even in that small area, it was seeing things it hadn’t seen before.” One of those things was an articulated bus. To a human driver, an articulated bus looks like a bus, just longer and bendier. But to an unacquainted AI, it’s an entirely new object that needs to have its form classified and its behaviors cataloged.
Regardless, things seem to progressing quickly on the AI side. Tesla, which has been gathering data over the past two years from the tens of thousands of cars equipped with its Autopilot feature, expects to roll out “full self-driving capability” within three to six months, according to CEO Elon Musk. That aggressive timeline is at least three to four years ahead of those of other major automakers, perhaps in part because Tesla has decided to equip its cars with radar and cameras in lieu of the more expensive (but also more precise) lidar. NuTonomy, which does use lidar, expects its service to roll out in select cities starting next year.
Full autonomy will likely become widely available when sensors become sufficiently inexpensive. Some companies are making strides in slashing costs. Early this year, Waymo, Alphabet’s self-driving car company, announced that they had reduced the price of their lidar units by 90%, a significant drop, but given that the originals cost $75,000, they’re still prohibitively expensive. Even if consumers are willing to pay almost $15,000 extra for a fully autonomous car, one of today’s lidar units would make up nearly half the cost. Solid-state lidar could cut costs dramatically. Quanergy, a start-up, last year announced a $250 lidar unit that they expect will go into manufacturing this year, though it’s not currently marketed as suitable for self-driving cars.
Barring a breakthrough, sensor costs will likely remain high for many more years. But that may not matter. The other way that self-driving cars could catch on is if none of us buy them in the first place.
21st Century Car Pooling
Halfway around the world from South Boston, in Singapore’s One North business district, nuTonomy’s self-driving cars have spent the last several months picking up passengers via the ride-hailing service Grab, Uber’s rival in Southeast Asia. As in Boston, there’s a human ready to take the wheel, but otherwise the cars operate on their own.
Such ride-sharing fleets could hasten the adoption of autonomous vehicles. “From a technology and economic perspective, it’s vastly more likely that autonomy will be used for mobility services,” Iagnemma says. It mostly comes down to economics. Private cars sit parked around 95% of the time. But for fleets operators like nuTonomy or Uber, the cars will be in use a large portion of every day, making it easier to justify the added costs.
Even if autonomous features cost tens of thousands of dollars per car, for transportation network companies, also known as TNCs, that’s cheaper than paying a person to sit behind the wheel. “For Uber, the number one cost is driver labor,” says Donna Chen, an assistant professor of engineering at the University of Virginia. In Los Angeles, for example, about two-thirds of the average fare goes to pay the driver. Uber drivers’ self-reported hourly rate is about $14–15 per hour, which means that $30,000 worth of sensors and hardware could be recouped in just 120 18-hour days. “TNCs have an enormous economic incentive to get the driver out of the vehicle and get electronics and tech in,” Tillemann says. “The driver is by far the most expensive part.”
Even with human drivers, Uber and the like have been relatively successful. Uber reported making 62 million trips last July, and Lyft reported 17 million last October. Still, those numbers pale in comparison to the 957 million trips Americans take every day in private vehicles. Cost is one factor holding TNCs back. At over $1 per mile, they’re significantly more expensive than driving your own car. There’s an easy way to make it cheaper, and that’s to split the cost between multiple passengers. But are Americans ready for that?
“It’s a cultural question,” Iagnemma says. “Will people want to share cars?” Currently, most people don’t appear interested. Only 15% of people have ever used a ride-sharing app and just under 10% of commuters carpool, but Iagnemma thinks a generational shift is underway. Many Millennials delayed their first car purchase, which gave Uber, Lyft, and the others an opening, and that familiarity could make the transition to shared autonomous vehicles easier.
The easiest way to speed adoption of ride sharing, of course, is to lower costs to the point where it doesn’t make sense not to use them. Intelligent routing algorithms will certainly help—forecasting demand based on factors like time of day and weather so that passengers don’t have to wait too long or go too far out of their way to pick up another rider. Simulations of such systems suggest that they could work at a surprising range of population densities, from build-up urban cores to more suburban neighborhoods.
Autonomy will also go a long way to driving down costs, but so, too, could electrification. “The costs of batteries have come down dramatically over the last ten years, and electric vehicles are just cheaper to maintain than standard cars,” Tillemann says.
So are EVs up to the challenge? I went to Amsterdam, home of the largest all-electric taxi fleet in the world, to find out.
Damp and brisk, the North Sea air fills my lungs with relief after six hours on a stale airplane. At the front of the taxi stand, a man beckons me to “pick any one you like.” Dozens of cabs are lined up. There are black Mercedes sedans, blue Sprinter vans, and sleek Teslas. I take the first Tesla, a white one.
The ride is remarkably quiet. If you’ve ever spent time in an electric vehicle, you know the sensation. Aside from the rush of the wind and the muffled hum of tires against pavement, there’s not much else to disturb the quiet.
The driver, G.M. Pierzad, has been piloting taxis to and from Schiphol Airport for 18 years and this Tesla Model S 85 for a little over two years. He splits it with another driver, which helps make the substantial lease and medallion payments a little more manageable. While the Tesla isn’t his favorite taxi—he prefers the Mercedes S-class—he clearly has gotten used to driving it, making full use of the car’s prodigious torque to squeeze into slots in Amsterdam’s busy rush hour traffic. For him, the car’s range isn’t quite enough, and he often stops at Schiphol’s Superchargers, which deliver up to 120 kW of electricity and can charge 75–80% of the battery capacity in 30 minutes. In a perfect world, he would only have to charge between shifts. “Maybe after one or two years it will change,” he says, which is when he’ll be due for a new taxi. A few years may be optimistic, but five to ten is more reasonable. By 2025, either the range problem will be solved or all taxis in the Amsterdam will face it—by that point, the city government will require the entire fleet to be zero-emissions.
After a brief nap at the hotel, I’m picked up by another white Tesla, this one driven by Erdal Ova, a 16 year veteran of the Schiphol taxi scene. Joining me is Gamis el Bouakili, director of Schiphol Taxi, one of the two companies that run the airport’s electric taxis. He manages a fleet of around 1,200 vehicles, including 101 Teslas.
Schiphol Taxi’s Teslas were delivered in September 2014, and the first few months were filled with headaches, el Bouakili says. There were no Superchargers available, so drivers had to rely on slower 22 kW charging stations and lower-powered public chargers. A full charge would take eight hours or more depending on which charger was available. In those first months, so many drivers with fares ran out of range that the airport and the taxi companies began requiring drivers to have a minimum 30–40% state of charge before they could enter the queue.
Today, with temperatures hovering in the low 40s, the battery’s charge is dropping quickly. Both the cabin and battery need to be warmed, sapping range. In the summer, air conditioning poses the same challenge.
Schiphol Taxi’s drivers have to account for dozens of variables as they manage their shifts. Like Pierzad, Ova is an owner-operator, meaning that every minute he’s not running a fare, he’s not making money. “You really have to plan your shift,” he says, especially if you don’t have access to Superchargers. “Without the Superchargers, it makes it harder to make a profit.” He and the other drivers charge up to three times per day, though he has made it through an entire shift without charging. (In fact, Ova is the most efficient electric driver at the airport, el Bouakili says, which entitled him to a trip to Tesla’s California headquarters.) On our tour, whenever we stopped at a location with a Supercharger, Ova plugged in, even if we were only there for five minutes.
Because repeatedly charging and discharging a lithium-ion battery’s full capacity can swiftly degrade it, Tesla recommends that drivers keep their cars 30–80% charged, only using the full range when they really need it. For the average owner, that’s a reasonable request. But taxi drivers need as much range as possible every day. In the beginning, Ova says he always charged the car to 90% full. Perhaps as a result, he estimates that his battery has lost maybe 2–3% of its capacity per year. Other drivers have likely experienced similar losses—Pierzad mentioned that his range has dropped, too. And perhaps as a result of that, Tesla recently pushed an update to the Schiphol fleet limiting their maximum charge to 80%.
Since Ova received his Tesla in late 2014, he’s logged over 90,000 miles in the car. It’s had some small problems—the door handles that magically extend when touched seem to get stuck with some frequency—but otherwise he’s only had to replace the tires and, less frequently, the brakes. Such reliability is not terribly surprising, given that an electric motor has one moving part whereas a traditional internal combustion engine has hundreds.
The next day, we make our way to an industrial park just south of Schiphol Airport. There, I meet up with Tofik Ohoudi, a business unit manager for BIOS-groep, a Dutch transportation company. He manages their fleet of blue Teslas that serve the airport. Unlike el Bouakili’s company, BIOS-groep owns the cars and employs their drivers directly. Supercharging—while certainly helpful—isn’t as vital as it is for Ova, Pierzad, and the other white Tesla drivers. When one of BIOS-groep’s cars is running low, the driver can simply drop it off at the depot and pick up another. Still, Ohoudi welcomed the recent addition of ten Superchargers for the airport taxis. “Every driver that starts a shift has to charge at least once, and every hour they have to charge costs us extra,” he says.
Ohoudi’s fleet has been similarly low-maintenance. (Again, the door handles were one of his biggest complaints.) That’s probably helped by the fact that, like many taxi operators, BIOS-groep had the power of the cars reduced. The detuning not only saves wear and tear—tires only have to be replaced every 30,000 miles or so, Ohoudi says, and he hasn’t had to replace the brakes yet—it also prevents the drivers from accelerating too rapidly, which can quickly deplete a battery’s charge.
And like el Bouakili, Ohoudi can work with his dispatchers to monitor the cars’ states of charge—Tesla built them a web app to keep an eye on the fleet—and recommend to the drivers when its time to give the car some juice.
The BIOS-groep depot is a massive open building with a half-wall running down the middle. On either side, a handful of cars are parked for a midday recharge at the 22-kW stations. But between 2 and 5 am, nearly all of the cars are here charging. “Drivers always start their shift with charged cars,” Ohoudi says. “We need the full capacity of the battery pack to do our job.”
Despite BIOS-groep’s relatively conservative usage, Tesla still had to replace all of the cars’ original batteries last year under warranty. “I think Tesla will change a lot of specifications and contracts with new owners because of the problems we’ve had,” Ohoudi says.
Despite the problems, for Ohoudi, the Teslas have become just another vehicle in his fleet. “For us, it’s a normal issue at this point,” he says of charging and the battery issues. “It’s not that exciting. We want to earn money with these cars, and we do.”
By 2025, when Amsterdam requires zero-emission taxis, it’s possible that it’ll be even easier for Ohoudi and his colleagues to earn money with EVs. Bloomberg New Energy Finance, a research organization, predicts that new long-range electric vehicles will cost the same up front as a gas-powered car somewhere between 2022–2024. “But it could well happen earlier,” says Ethan Zindler, an analyst with BNEF.
Putting the Pieces Together
The race is on, then. If autonomous vehicles and EVs are widely available by the early 2020s, which appears to be the case, then our transportation system is due for an upheaval not seen since the Model T. Exactly how it all shakes out is anyone’s guess, but transportation experts have some well-informed predictions.
The biggest factor will undoubtedly be cost. If the services aren’t cheap enough, people won’t use them. Private fossil-fuel cars cost somewhere between 50 cents and one dollar per mile to own and operate. Today, ride sharing services like Uber generally cost more than that, with most of that pay going to the driver.
Autonomy would reduce those costs considerably, making ride sharing more accessible. In turn, ride sharing could bring self-driving cars to the masses faster than if everyone waited to buy one themselves. “The initial prices will stay high, so the fleet operators will get first dibs on most of those vehicles,” says Kara Kockelman, a professor of engineering at the University of Texas. “Maybe five or ten years later you’ll see more private ownership.”
Transportation network companies, says Iagnemma, the nuTonomy CEO, “can more easily amortize the technology costs.” That’s in part because of a quirk in the way we humans value our own time. We generally don’t consider time spent driving to be a cost like gas or insurance, which means that, to individuals, autonomous features will feel like pricey luxury at first. But for companies like Uber, autonomy will be a bargain compared with the cost of paying human drivers.
Autonomous cars would also solve a problem that will eventually limit the growth of mobility services. “There are not enough people around,” says Emilio Frazzoli, an aerospace engineer at MIT and cofounder of nuTonomy. “If you want to provide this kind of service for everyone in the world, then one out of four or five or six people must be a driver. That is never going to happen. People still will need teachers, doctors, policemen, and firemen; some people will still need to be kids.”
If autonomy does catch on, and transportation network companies do buy them in big numbers, then the final piece of the “heavenly” puzzle will likely fall into place. “If we’re looking at a system that is dominated by TNCs, and fuel is a substantial portion, they have a very serious incentive to push toward electrification,” Tillemann says.
Fleets of autonomous cars would simplify the adoption of electric vehicles in myriad ways. For one, they would eliminate so-called range anxiety, where people fear running out of charge. The cars could charge themselves. In fact, most experts I spoke with agreed that autonomous recharging would be simpler and safer than robotic refueling of gasoline. And the network operator would only send a car that they know could get a passenger to her destination. “Suddenly, she doesn’t have to worry about charging, she doesn’t have to worry about anything,” Kockelman says. “That’s one of the great benefits of a shared fleet.”
Shared electric fleets would also spread the costs of batteries—which can currently be up to 25% of the price of an EV—out over hundreds, even thousands of customers. “If you don’t have to buy the electric car, but you can essentially use one whenever you need it, then the capital cost of buying the battery goes away,” Frazzoli says. “Effectively, you are sharing it with other people.”
Shared autonomous electric vehicles could also open up a new avenue to meeting emissions targets to stave off the worst of climate change. Frazzoli and his colleagues recently published a paper showing that intelligent ride-sharing algorithms could trim the New York City taxi fleet down from 14,000 vehicles to just 3,000. Elsewhere, ride sharing would have a similar effect, with three to four cars being taken off the road for every shared vehicle added, Iagnemma says.
And if those shared autonomous vehicles are powered by electricity—which seems likely—and if that electricity is produced by carbon-free sources—which isn’t as far-fetched as it sounds—then meeting an emissions target by 2050 could be possible.
To keep global warming to 3.6˚ F, the upper limit of what scientists consider safe, we’ll need somewhere around 410 million electric cars on the road by 2050, according to the Global Calculator tool developed by UK’s former Department of Energy and Climate Change. That assumes that we’ll keep buying and driving cars like we do today. But if shared autonomous electric vehicles arrive on time and as expected, it could reduce the number of cars we need to keep people moving and, in turn, make transportation emissions one of the easiest climate change challenges to surmount.
“If we adopt policies that encourage a shared autonomous electric transportation system,” Tillemann says, “then it can make an enormous difference and much faster than most people expect.”
With reporting by Ana Aceves.