SAN FRANCISCO — Silicon Valley is the land of the beta test, the constant tweak, where companies habitually release products still in development to see how they work in the hands of consumers.
Last week, that iterative approach, so ubiquitous in software, entered a new realm when Uber announced that it would begin testing a fleet of 100 self-driving cars for hire in Pittsburgh by the end of the month.
The move means that the streets of a large American city, one that gets an average of 41 inches of annual snowfall and has more than 400 bridges, will become the company’s laboratory. And the test subjects will be real people who summon the vehicles, some weighing more than two tons with turbocharged engines, with their smartphones.
Autonomous vehicles represent the future of transportation, industry officials say, with the potential to cut down on accidents and fill in gaps in public transportation. Perfecting the technology is a top priority for companies such as Google and Apple and automakers such as General Motors. But the dream of cars that can reliably and safely navigate urban areas on their own is thought to still be years away. The technology also suffered a significant setback only recently, when a driver of a semiautomated Tesla got into a fatal crash.
Even experts in automated driving were therefore surprised by Uber’s move, which will mark the first time everyday commuters will be able to hail a ride in a driverless car.
“They’re bringing their customers into the development process,” said Pete Koomen, founder and president of Optimizely, a start-up that tests alternatives for websites, called A/B tests. “The customers understand that their goal is to create the best product possible, and they need their help in doing this.”
To help allay concerns and to comply to with state law, which requires a driver behind the wheel, Uber will have two trained safety drivers on each ride.
Earlier this month, Uber announced it was acquiring Otto, a tech start-up that makes self-driving trucks, in a move that Uber said gives it “one of the strongest autonomous engineering groups in the world.” In Pittsburgh, Uber has partnered with Carnegie Mellon University’s Robotics Institute, which has three decades of building automation technology.
The company points out that human error is the cause behind the overwhelming majority of traffic deaths, with motor-vehicle accidents the leading cause of death for those under 25.
“This is a tragedy that self-driving technology can help solve,” the company said.
Even with safety drivers, Uber’s wager is risky because the project will allow vastly more people to see the technology’s flaws, said Bryant Walker Smith, a law professor at the University of South Carolina who researches self-driving cars.
“This is going to expose the technology to the public but also will give them a much more intimate view of its mistakes,” he said. “They’ll see in a very direct sense that these are technologies in progress.”
He said he expects that the program will engender trust and that riders who experience the beta test are likely to walk away with a positive idea of automation. “The astounding thing is that people think that these are terrifying until they get into a vehicle,” Walker Smith said. “Then, a mile later, they have total and complete confidence in these systems.”
But for regulators to truly know whether a self-driving car is safe, they will need reams of data that come from hundreds of millions of miles of testing, he said.
By that all-important metric, self-driving cars are a long way off. As of June, Google’s fleet of roughly 50 autonomous vehicles had driven 1.7 million miles without a fatal accident, the company said. According to the Rand Corp. and other experts, self-driving cars need to be tested in real-world conditions for hundreds of millions more miles — at least — to be considered safe.
These test trips enable the automation software to acquire the huge amount of data necessary to detect the difference between a puddle and pothole — something that today is still challenging, experts said. Uber declined to share how many miles its self-driving program had logged.
Despite their safety measures, Uber, Google, Tesla and other automakers that are testing or deploying these systems should be prepared for the worst, too, Walker Smith said. They should have a plan for addressing the first serious injury or fatality that might take place, he said.
In Silicon Valley, companies often use beta tests to get useful feedback before products are ready for prime time. Product development is more efficient when you can iterate continuously, rather than expend a huge effort to mass-produce something that may be a flop.
“There’s always a trade-off between wanting to release early to guide your development and releasing too early when it’s not ready,” Koomen said.
But Uber’s plan is perhaps the most significant example yet of the digital ethos of continuous testing now moving into the physical world.
In an interview, Pittsburgh Mayor William Peduto said Uber chief executive Travis Kalanick had approached him about the project a year ago. In their first meeting, Kalanick referred to his plans for “the Pittsburgh Project,” a term that took the mayor by surprise. Kalanick said it was like the Manhattan Project, and the idea was to use the company’s resources to turn the city into a hub for autonomous cars.
Initially, the self-driving pilot was slated for a November rollout. About two weeks ago, the mayor got a call from Kalanick saying that the timetable had moved more quickly and he was ready to launch. On the day of the announcement, Kalanick came to Pittsburgh and took the mayor for a ride in a self-driving car.
“It was a steady ride home. There was no time I was fearful or worried,” he said, adding that the safety driver took over the wheel twice during his trip. “I’m more worried when I’m on the road with an 18-year-old who is learning how to drive.”
While the mayor may not be worried, some of his constituents are.
Eric Heyl, a columnist for the Pittsburgh Tribune-Review, wrote in a recent column that someone should “slam the brakes on this plan.” The problem isn’t just the self-driving cars — it’s Pittsburgh drivers, who, he wrote, “fail to properly merge, yield and stop.”