Skip to main content Skip to secondary navigation
Main content start

Traveling in the age of driverless cars

A team of Stanford researchers explain why completing the last few miles to our self-driving future is “devilishly difficult.”

Shelley, an autonomous Audi, at Stanford’s VAIL facility. | Photo by Toni Bird

Shelley, an autonomous Audi, at Stanford’s VAIL facility. | Photo by Toni Bird

Stephen Zoepf, executive director of the Center for Automotive Research at Stanford, has a wager with a friend over the future of driverless cars.

His friend is the bullish one. He’s betting that by May 27, 2024, a commercially available autonomous car will be able to transport Zoepf from a Las Vegas Strip hotel to the city’s downtown with no intervention from a human driver.

If his friend wins, Zoepf will donate $500 to Mothers Against Drunk Driving. If Zoepf wins because the driverless car isn’t yet road-ready, the friend will donate $500 to Mercy Medical Angels. (Oh, and the loser pays for the trip to Las Vegas.)

Four years after making their predictions, both men say they’re more firmly entrenched in their views than ever.

“We are 80 to 90 percent on our way to an automated commercial vehicle,” says Zoepf, still in his biking shoes and shorts as he sits down at Stanford’s Volkswagen Automotive Innovation Lab. “But the last 10 percent is pretty hard.”

That part — that last 10 percent — is attracting attention from researchers, investors, businesses and consumers, and it’s become a focal point of extensive collaboration for dozens of Stanford professors, students and affiliates. It is still not clear whether we are headed for an efficiency-oriented network of shared autonomous electric vehicles that look like “shoeboxes on wheels scooting around the city” — as Zoepf says many people envision — or cars so luxurious they can transport us while we work out on built-in exercise equipment or relax with guided meditation. Most likely it will be a combination of the two.

Bold predictions abound. Babies born today won’t ever need a driver’s license. In two years, we’ll have 10 million self-driving cars on the road. A 2017 report examining car ownership and technology by RethinkX, an independent think tank that studies technology disruption, says we are “on the verge of an explosive transformation.” To wit:

  • Only 20 percent of Americans will own a car in 15 years.
  • Passenger vehicles on American roads will drop from 247 million in 2020 to 44 million in 2030.
  • Driverless technology will transform many, if not most, modes of transportation: trucks, tractors, ships, forklifts, trains, construction equipment.
  • Transportation costs will drop to about 20 cents a mile, at which point everyone will have affordable access to road travel. (That’s about a threefold decline, according to a 2017 AAA report.)

Disruption may be too mild a descriptor for what is happening to our means of transportation. How we get from here to there is not merely a convenience, but rather has the potential to affect poverty levels, social standing, our daily human interactions, cultural norms and even life span, according to Stanford scientists and researchers.

On the one hand, self-driving cars could democratize transportation, making independent travel possible for many who lack it, including people who are blind, disabled, young, old or poor. Prognosticators also expect reductions in pollution, traffic, collisions and the cost of getting around, and an increase in green space as the demand for parking lots declines.

But skeptics worry that the adoption of driverless cars will eliminate too many jobs and give hackers a new way to attack, even possibly turning cars into lethal weapons. As the debate unfolds, autonomous vehicles are hitting the road in droves. Ready or not, say experts, here they come.

At the Garage

Dotted with start-ups tackling the business of autonomous vehicles from every angle, Silicon Valley has been dubbed the new Detroit, which might make Stanford’s auto lab the downtown.

The Center for Automotive Research at Stanford (CARS), launched in 2008, was co-founded by mechanical engineering professor Chris Gerdes, computer scientist Sebastian Thrun and the late professor of communication Clifford Nass. It now serves as a hub for collaboration across Stanford’s seven schools. Through the center and at its research site, the Volkswagen Automotive Innovation Lab (VAIL), aerospace and mechanical engineers, philosophers, artificial intelligence experts, neuroscientists, psychologists, designers and ethicists come together to study and help create the vehicles and transportation systems of the future.

On the north side of campus amid coast live oaks, the VAIL facility is both serene and futuristic. From the worn wooden picnic bench under the trees where he sometimes sits to work, Gerdes, CARS’s director, can look into the giant windows at the front of the lab and see the lineup of vehicles — Shelley, the autonomous white Audi TT-S; MARTY, the self-driving DeLorean race car; and Stanford’s award-winning solar car, Sundae, built and driven by a team of students. Each occupies a bay. Nondescript customized test cars come and go from outdoor parking spaces. The fleet is always growing, as projects expand and multiply.

In keeping with its collaborative approach to development, the center partners with more than 40 corporate affiliates, including a who’s who of car companies — among them Ford, General Motors, Honda, Mercedes-Benz, Nissan, Toyota and Volkswagen. Each affiliate pays $32,000 per year to support and team up with students, faculty and researchers from the Law School, the Graduate School of Business and the School of Engineering.

Zoepf says his goal is to make transportation “truly sustainable.”

“That is something I tell all my students. We can’t do damage to the environment or economic stability. We aren’t just providing transportation for the upper 5 percent of earners who can buy new autonomous cars. We are providing multiple transportation solutions for everybody.”

The Theater of the Car

In the back room of VAIL is a door that opens to the simulation theater, home to a matte black Toyota Avalon surrounded by 240-degree floor-to-ceiling screens. Most of the time, the theater is run by students who program a wall of computers that dictate what happens. It is here that undergraduates, graduate students and volunteers test out scenarios that an autonomous vehicle might encounter on the road.

From the simulator’s passenger seat, the view is expansive and realistic. As we begin to drive, cars speed past in the opposite lanes. High rises, trees and street signs mark the roadway. The brain knows this is virtual reality, but the experience is immersive. Surround sensors measure the car and the humans’ every response, collecting data as we ride.

For Wendy Ju, ’97, PhD ’08, who led interaction design research at Stanford’s Center for Design Research until she moved to Cornell Tech in January, the autonomous car presents a perfect storm of research challenges related to human-machine interaction. How quickly can the passenger take over driving in an emergency? What types of distractions cause the most problems if the rider is required to react? Why do people get so sleepy in the simulator, and what can designers do to counteract inattention?

“There is a last mile with driverless cars,” Ju tells me. “The biggest challenge is that we don’t know yet what it means to be consumer-ready. We need to understand how people are going to interact around autonomous vehicles so that they are safe and effective.”

One of Ju’s most notable experiments involved the Ghost Driver, a customized Volkswagen eGolf in which a human driver was disguised with a costume that made it appear as if there were nobody in the driver’s seat. The Wizard of Oz experiment, as Ju and her co-authors called it, allowed them to videotape pedestrians as they interacted with the seemingly driverless cars. The researchers also administered follow-up questionnaires to glean reactions from walkers and cyclists who interacted with the car.

Even in Silicon Valley, pedestrians at a four-way stop do a double take when they realize there is no one in the driver’s seat. A few whip out cell phones to snap a photo. When the car behaves as it should, stopping fully and waiting for pedestrians to cross, all goes smoothly. When the car inches forward as the pedestrian enters the street, confusion ensues — the walker might stop and start, pass behind the car, or make an otherwise unpredictable move. The research shows not only how people respond to autonomous driving, but also how engineers must make cars “understand” the people-rich contexts they encounter.

Ju says she finds autonomous cars “really, really exciting,” but in the final analysis, she adds, “I predict most people will find the actual experience of being in an autonomous vehicle to be a yawn.”

Would You Ride in a Driverless Car?

You may have already shared the road with one of Waymo’s driverless cars, which have traveled more than 4 million miles on city roads. Uber’s self-driving truck made its first deliveries without a human driver in 2016. Also that year, Uber was the first to put driverless cars to passenger road tests in an experiment in Pittsburgh. Tesla’s Autopilot technology now offers lane centering, enhanced cruise control, self-parking and the ability to summon the car from its parking spot. And multiple carmakers say they’ll have self-driving cars on the road by 2021, the year autonomous vehicles are estimated to be a $42 billion industry.

Still, the public is expressing uncertainty about the concept and a deep skepticism about the trustworthiness of completely driverless driving. A 2017 Pew Research Center poll reported that nearly six in 10 Americans say they would not want to ride in a driverless vehicle, and 87 percent of Americans believe every driverless car should have a human capable of taking over in an emergency.

“It takes very little to make someone uncomfortable that they are not driving,” Ju says. “I call it the husband factor. If you were in the passenger seat and your husband were driving, you could say, ‘Honey, look out for that other car.’ It isn’t that I don’t think he knows how to drive. It is that I still get to express my opinion. You don’t get that in an autonomous vehicle.”

The biggest challenge here is that driving is so open-ended, says Gerdes. Most of us have experienced enough technological malfunctions, glitches and bugs that it isn’t difficult to contemplate them in the context of an autonomous vehicle. What if the car makes a bad decision and hits a pedestrian? What if, as happened recently, a self-driving car gets confused because there is graffiti on a stop sign and it can no longer interpret the sign? What if hackers break into the satellite system that downloads updates to your car?

“You can’t program for everything,” says Gerdes, who drives an all-electric black Chevy Bolt. (He is quick to add that he also loves to get behind the wheel of the self-driving DeLorean his team uses to test high-speed autonomous maneuvers.) “It is an infinite number of possibilities. There is also no real answer as to when you are done with development,” he says.

What is the acceptable level of risk for autonomous vehicles, and who is responsible for bad outcomes? There isn’t yet consensus on those questions. Assessing human tolerance for error and accepting the outcome of a car programmed to act in a way that may lead to injury are just a couple of the issues facing researchers today.

Still, for some riders, the advantages outweigh their fears, and therein lies promise.

One night last fall, a crowded room of aspiring entrepreneurs gathered in a basement classroom on campus for BUS174: The Business of Self-Driving Cars, a course in which students build out business plans and present them to a judging panel. The guest speaker was Oliver Cameron, CEO of Voyage, a Silicon Valley company that introduced driverless taxis to a senior community in San Jose called the Villages. A video showed a delighted elderly woman who is legally blind using the Voyage app to summon a taxi that picks her up and gently transports her to her destination within the 15-mile campus of 4,000 residents.

“People associated with the adoption of technology were very skeptical about this trial,” says Cameron, dressed in the obligatory Valley hoodie for his presentation, but the residents greeted it with enthusiasm. “We had people coming to us with vision problems, Parkinson’s and mobility issues,” Cameron says. “They want to have fun and go out like anyone else. It has been a community that wanted to adopt it. The true test will be breaking out of those communities into the rest of the world.”

When The Going Gets Complicated So Do The Ethics

Although driverless vehicles are expected to greatly decrease traffic fatalities, they will sometimes have to decide between bad options. Would your car swerve to miss a cyclist at the risk of hitting another vehicle head-on? Would it protect its own occupant at a greater risk to others?

Not surprisingly, respondents to a series of 2016 surveys reported in the journal Science said that — in concept — they believe autonomous cars should sacrifice their passengers’ well-being for the greater good. But they personally would not buy or ride in such a vehicle.

Jason Millar, a postdoctoral research fellow in engineering and ethics, is one of a dozen scientists and researchers from across Stanford tackling some of the prickliest questions keeping automakers up at night. 

“It is one thing to say that a machine is working better than the average human,” Millar says, “but that is based on what it means to be better. The set of criteria you use is negotiable. It can be contentious. That is where some of the interesting ethical questions arise.”

Millar wants to equip engineers to make some of the most important decisions about safety in the design room. He predicts engineering ethics will become an important new field. “We need to get engineers solving these problems who are attuned to human ethics and human knowledge,” he says.

Such cross-pollination is part of what brought Mykel Kochenderfer, ’02, MS ’03, an assistant professor of aeronautics and astronautics, to the autonomous vehicle party. As director of the Stanford Intelligent Systems Laboratory, he develops advanced algorithms and analytical methods for use in air traffic control, unmanned aircraft and other applications in which decisions must be made in uncertain, dynamic environments while maintaining safety and efficiency. The relevance to Gerdes’s work became clear when Kochenderfer joined Stanford in 2013.

Kochenderfer notes that bringing autonomous transportation to places outside the Bay Area — which benefits from “pristine” weather and a culture of innovation — will be a challenge. On a recent trip to Boston, he found himself at an intersection that took all his cognitive abilities to figure out how to navigate safely. In that moment, he could not imagine a driverless car being able to handle it.

“A lot of what we bring to autonomous vehicles comes from aircraft avoidance systems,” he explains. “In that context, you can be really, really safe if you maneuver when you see another aircraft.” But on the road, the proximity of other vehicles is a given. For example, on a two-lane road, an approaching truck can plow into a car in less than a second.

“You have to reason about the likelihood that they will invade your lane,” he says. “There is nothing you can really do about it other than [not] leave your driveway.” After a long pause, Kochenderfer adds, “I’d agree on the assessment that the last 10 percent is the hardest.”

Regulators — another key part of the safety puzzle — also have a ways to go in adapting laws and policy to autonomous driving. As of early January, 27 states had passed legislation or issued executive orders related to autonomous vehicles, according to the National Conference of State Legislators. While many of these laws have eased the deployment of self-driving vehicles, legislatures have yet to resolve how they will handle fundamental issues such as liability, insurance and safety inspections. Industry observers describe “a Wild West,” where adequate safety protections for consumers are not yet in place and where regulatory purview over the industry has yet to be determined. (At press time, a 2017 House bill seeking to consolidate power under the U.S. National Highway Traffic Safety Administration had yet to go before the full Senate.)

For his part, Zoepf says he’s excited about how fast policy is moving. “Before 2013, the U.S. Department of Transportation had not uttered the words ‘autonomous vehicle,’” he says. Now there are bills in the House and the Senate that include provisions allowing for up to 100,000 vehicles per manufacturer per year on public roads.

The Economics of a Transformation

Imagine you summon a ride using an app on your phone and an electric pod vehicle arrives, the door pops open, and you get into your compartment. (This part has already occurred in road tests.)

“I can play my Dungeons and Dragons, bring dumbbells and work out, or get paired with people who want to do business with me,” says Gerdes.

Not only could your vehicle get you where you are going, but it also could potentially pick up your dry cleaning while you are at work or the gym, drop your teens at the mall, transport many other passengers, and be back to collect you at the end of the day.

While experts can’t fully predict the new economics of an automated, largely electric transportation system, there is consensus that for the rider, shared mobility will mean savings and a democratization of transportation. As commuting via driverless vehicle becomes cheaper, consumers will no longer need to hang on to their individual cars.

“Lack of transportation is the single biggest barrier to climbing out of poverty,” Gerdes says.

As the costs of transportation decline, many more people will be able to commute affordably to work and fulfill the responsibilities of daily life, from grocery shopping and dropping the kids at school to visiting the doctor or showing up for jury duty.

Driverless cars are also expected to have profound effects on a broad range of industries, from insurance companies to fee-based parking lots, gas stations to car mechanics, auto parts stores to professional drivers.

The trucking industry could be hit hardest. According to the American Trucking Association, some 3.5 million truck drivers work in the United States. Automation presents a clear threat to that livelihood: As of 2014, truck, delivery and tractor driving was the most common job in 29 states.

The market for collision repair services would likely decrease, but the need for vehicle maintenance might grow as autonomous fleets spend more time on the roads than today’s personally owned vehicles.

Modern vehicles are also expected to collect loads of timely data for advertisers. Starbucks on your way to work? Need to stop at the neighborhood drugstore? Your car will connect you to commerce. And a multitude of companies stand to benefit from follow-on innovations, be they data logs characterizing riders’ routines or creations for use in transit. (Infinity pool for the car, anyone?)

So what about those endless parking lots where your car sits most of the time? They will become obsolete. Today, cars are parked 95 percent of the time, according to Paul Barter, a transportation policy scholar from the National University of Singapore who writes the blog Reinventing Urban Transportation. Already, some cities are creating parklets of green space instead.

In the short term, the biggest winners are likely to be ride-sharing companies like Uber and Lyft, which will be able to stop paying drivers. The losers will be those who make a living driving.

But I Love My Car

No one knows how long it might take for humans to fully accept the idea that they don’t need a car, or even a steering wheel. Our attachment to our automobiles is historic and powerful. Cars reflect their owners’ taste, wealth, lifestyle and self-perception. A symbol of freedom and independence, the car represents much more than a necessity to many Americans.

“We love our cars in the United States,” says Millar, who studies people and their automobiles as an ethicist and an engineer, but also as a car enthusiast — albeit one born in Canada. “It will be difficult to eliminate the human driver from the equation for sheer love of getting behind the wheel. It will be difficult to force people to give up their cars.”

Reilly Brennan, a Stanford lecturer and a founding partner at Trucks, a venture capital fund that invests in transportation start-ups, agrees. As co-teachers for the Stanford course Judging Historical Significance: The Automobile, Brennan and professor of mechanical engineering David Kelley, MS ’78, explore the importance of the car as an object unlike any other.

“Part of the idea of freedom in America is defined in some part by your ability to move,” says Brennan. “Not just to take a trip, but we are talking about your economic independence and your ability to pick up and go West … For the past 70 years, the car has been the physical embodiment of independence.”

But the vision of freedom will adapt, Brennan says. Take, for example, the great American road trip, a mainstay of family vacations. Will autonomous vehicles put an end to such adventures?

Far from it, say those in the industry. New autonomous features will enable riders in driverless cars, minivans and RVs to gaze out the window and listen to an audio tour guide that comes with their new vehicle. Gone will be the days of dangerously nodding off, or negotiating road hazards. One day, the whole family will be able to watch a movie or play cards. Or, for that matter, sleep.

Overcoming the "Last 10 Percent"

Addressing the last 10 percent of the challenges on the way to our self-driving future is “devilishly difficult,” Zoepf admits. He and his colleagues are not anxious to make exact predictions of how the driverless car phenomenon will unfold. What they will say is that over time, dependence on individually owned cars will diminish, and autonomous cars will continue to become more commonplace on the highways. Within limited localities, testing of fully automated vehicles will continue and expand. Some areas may even forbid manually driven cars for a period — but these will be few and far between. In industries such as trucking, construction and agriculture, automation will become prevalent. In most places in the United States, the introduction of widespread automation will take place slowly, through increasingly autonomous features in traditional cars.

Zoepf is asked almost daily to foretell the future of autonomous vehicles. His favorite response comes from a presentation slide, depicted below, that he uses to show how complex the road to full autonomy is, and how oversimplified many have made it sound. If only it were that simple to achieve a utopian autonomous future, he says, laughing.