On a Sunny afternoon in early October, I drove across a parking lot in Ann Arbor, Mich., and down a sloping road to a second, private lot. A fence lining the lot’s perimeter was covered in black fabric, as if to deter snoops. Behind it was a 32-acre Potemkin village. There were paved roads with names and signs: Liberty Street, Main Street, Wolverine Avenue. There was a traffic roundabout, a covered underpass and a railroad crossing. There were cosmetic props, too — newspaper boxes, sidewalk benches, a row of fake storefronts.
Mcity, as the facility is known, was built by the University of Michigan as a testing ground for automakers, including Ford, which has been experimenting with self-driving cars here for about two years. I parked my rental and climbed into the back of a white Ford Fusion with four spinning Lidar sensors on the roof, each roughly the size of a water glass, and a rack of high-performance computers in the trunk. Next to me sat Randy Visintainer, Ford’s director of autonomous vehicle development. Jakob Hoellerbauer, a young Ford engineer, took the driver’s seat. Wayne Williams, a Ford research scientist, sat beside him in the passenger seat typing on a laptop. We put on our seatbelts, and Hoellerbauer hit the ignition and pulled into the road. A beep came from the dashboard, indicating that the car was ready to engage self-driving mode. Hoellerbauer pushed a button on the steering wheel and took his hands away.
The car glided along Main Street to a four-way stop, waited for another car (driven by a researcher) to clear the intersection, then made a gentle right turn. Autonomous cars are programmed to drive conservatively, and any time our Fusion sensed an approaching object — like a Ford employee, role-playing an oblivious pedestrian — it slowed down until it was sure the path was clear, then proceeded cautiously. It felt like being chauffeured by an elderly ghost.
Two minutes into the drive, the car pulled around a traffic circle and up to a blinking red light at an empty intersection. It stopped. A few seconds passed. Perhaps, I thought, it was just being characteristically risk-averse. Then a few more seconds passed, and the engineers started to look nervous. Williams squinted at his laptop.
“So right now, we’re not sure if the intersection is clear, so we’re being very cautious,” Williams said, speaking for the car. “It’s expecting …” He trailed off and turned to Hoellerbauer. “Why don’t you just overtake it?” Hoellerbauer grabbed the wheel and tapped the gas pedal, overriding the self-driving system and manually steering us through the intersection.
“We didn’t see that happen this morning,” Williams explained, by way of an apology. We took another lap around Mcity, and when we reached the intersection again, the car did the exact same thing.
Self-driving cars raise the specter of horrific malfunctions — a code glitch that sends a car careering off a cliff, a remote hack that disables the brakes — but perfectly innocuous problems like the one we experienced will be far more common, and no less an impediment to road-readiness. There are dozens of autonomous vehicles being tested all over the world — 43 companies are testing 295 different self-driving vehicles in California alone — and not one of them has a clean record. A self-driving Uber vehicle zoomed through a red light in San Francisco last year. The sensors on cars used by Waymo, the self-driving-car division of Alphabet, Google’s corporate parent, have struggled in heavy rain and snow. I’ve had half a dozen self-driving-car experiences in my life, including a white-knuckle trip down Manhattan’s West Side Highway in a Tesla Model X whose owner had set it to “autopilot” mode, and I wouldn’t describe any of them as relaxing.
How quickly these rides will improve depends on whom you ask. Silicon Valley futurists, in typically sanguine fashion, predict that we are months, not years, away from a Cambrian explosion of autonomous cars. Waymo has a pilot program for its self-driving vehicles already underway in Arizona and says it’s “really close” to being fully operational. Lyft, which has partnered with a start-up called Drive.ai, is aiming to have a fleet of self-driving cars operating in the Bay Area by the end of the year. And Elon Musk, the founder of Tesla, which has offered semiautonomous features since 2015, has suggested that its cars could be fully autonomous as soon as 2019.
But that’s California, where optimism is the coin of the realm. In Detroit, the timeline looks slower and more incremental. Ford, in particular, believes that the first generation of driver less cars will be limited, capable of travelling only in commercial fleets inside carefully plotted urban areas. Other cars will simply get smarter without being autonomous, with features like collision prevention and self-parking becoming more common. Self-driving technology will eventually be more sophisticated and will one day be capable of full door-to-door autonomy in every possible area and condition, but as Ford sees it, that’s not going to happen overnight, or even very soon.
Despite its pragmatic restraint, Ford has invested hundreds of millions in self-driving cars and is making surprising progress. In a report earlier this year, Navigant Research placed Ford at the top of its leader board, ahead of tech companies like Waymo and Uber as well as auto rivals like GM and Toyota, based on Ford’s advanced manufacturing capabilities coupled with its strides in software development. The report generated headlines like, “Detroit Is Kicking Silicon Valley’s Ass in the Race to Build Self-Driving Cars,” and delivered a morale boost to Ford, which showed up late but finally felt as if it had a chance.
Ford’s big, blinking target is 2021 — the year it hopes to release a vehicle that meets the Society of Automotive Engineers International’s definition of Level 4 autonomy (no human operator required in the area and conditions it’s programmed for). It’s a tough deadline for a company whose culture might be the exact opposite of Silicon Valley’s, where companies release half-finished “minimum viable products” and abide by maxims like “move fast and break things.” Before Ford — or any conventional automaker — produces anything, each part has to pass a grueling battery of tests and certifications. There is a reason “automotive grade” has become a synonym for “reliable.” “Our vehicles have to be a trusted product,” Chris Brewer, Ford’s chief engineer for autonomous vehicles, told me. “That is a little more important than ‘Did my phone freeze or not?’ ”
Americans are wary of driverless cars — 56 percent, according to the Pew Research Center, would prefer not to ride in one — and when I talked to Brewer, it occurred to me that some part of that hesitation might stem from who we assume will be producing them: Silicon Valley tech giants, the same stateless behemoths that have spent the last few decades barging into old-line industries like the Kool-Aid Man, destroying working-class jobs and leaving behind cold, modern efficiency. But maybe these skeptics could be persuaded to trust Detroit. After all, Brewer is right — self-driving cars aren’t smartphones. They’re two-ton projectiles that take your parents to the grocery store and your kids to soccer practice, that will need to make billions of computational decisions per second while moving at 65 miles per hour, that contain within them the power to extinguish human life. You kind of want them to take a while.