There is a man in a straw hat who earns a paycheck here by crossing the street over and over again. Sometimes he stares down at his cellphone, often he wanders outside of the crosswalk, and he’s also known to ignore the traffic signal. What he does not do is pay any attention to the cars that barrel through the intersection.
If he did, he would notice that they have no drivers.
This carefully shrouded 91-acre preserve in the Central Valley is the testing ground for the driverless cars being developed by Waymo, the autonomous-car company that was a division of Google before being spun off last year. Waymo parted the gates of the tall chain-link fence, covered from top to bottom in opaque black plastic, to welcome a busload of reporters last week because they know their challenge is bigger than just developing the “perfect” driverless car.
“There’s some level of discomfort at being in a fully driverless car,” Waymo chief executive John Krafcik said. “We’ve read most of the [surveys], and for the most part, they say that about half of drivers are uncomfortable with this idea. And when we look at that, we say, ‘That’s cool, about half of users are already comfortable with the idea of going driverless.’ We actually see that as a reasonable starting place.”
The starting place may be a bit more dismal than that. An AAA survey this year found that three-quarters of people said they would be afraid to set foot in a driverless car, and more than half of drivers wouldn’t even want to share the road with one.
The folks at Waymo, and the multitude of other companies eager to put autonomous cars on the road, point out that once people experience the cars, they generally become converts to the technology.
Waymo says its driverless cars have driven 3.5 million miles in 22 test cities, racked up 2.5 billion miles in simulated tests last year alone, and have been challenged by about 20,000 what-if scenarios they might encounter on highways and city streets.
The company sent a blind man out alone in one of its cars in Austin, but much of the what-if tests (call it “courting disaster” experimentation) were carried out on this former Air Force base and at an earlier facility in Google headquarters 120 miles away in Mountain View, California.
Since the Austin test in 2016, the Waymo vehicle has evolved from a gumdrop-like looking car into a Chrysler Pacifica minivan that looks pretty normal, save for a dome on the roof about the size of an old-style police cruiser’s flashing light.
(For driverless-car aficionados, this is a Level 4 autonomous car. For the uninitiated who may harbor trepidation about a driverless future, that means that the cars have the ability to come to a safe stop, should anything go wrong, without driver intervention.)
Layers of detection systems collect far more data, from 360 degrees around the car, than a human driver could absorb.
Lidar detectors, which use a pulsed laser to measure the distance between things, are mounted on the front and rear bumpers and just ahead of the side-view mirrors on either side. It’s also in the dome, along with a high-resolution camera with a 360-degree field of view. Because it’s a color camera, things like traffic lights, yellow school buses, construction-zone cones and flashing lights on emergency vehicles show up. That’s supplemented with radar sensors on the four corners and top of the car.
Together, they picked up the man in the straw hat, several passing bicyclists, a row of traffic cones and changing traffic lights. The van also successfully negotiated a roundabout and dealt with three what-if scenarios during media day at the test sight.
The car also can hear.
The computer, and the backup computer that monitors it and takes over if anything goes wrong, has software that Waymo says “can model, predict and understand the intent of each object in the road.” Pedestrians, for example, move slower than cyclists, motorcyclists or cars, but they can change direction a lot more quickly. Then there is the computer’s “planner.”
“If your software perceives that an adjacent lane ahead is closed due to construction, and predicts that a cyclist in that lane will move over, our planner can make the decision to slow down or make room for the cyclist well ahead of time,” according to the Waymo manual distributed to reporters.
Krafcik and his crew, who work behind the opaque screens at the Old Castle air base, an airfield southeast of San Francisco which once dispatched B-52s to carpet-bomb Vietnam, are vague about when and how their cars first will see genuine action.
“It’s fair to say we are really close,” Krafcik said. “We’re not going to give you a specific date. We’re going to do it when we’re convinced that we’re ready.”
They say the details are fairly minor, things like helping an empty car to better understand just where it should pick up passengers, or the fact that a place the car selected for drop-off meant stepping into a field of unwelcoming cactuses.
Unlike the big-name automakers, Waymo nee Google decided to leap directly to fully driverless cars, while companies that have cars to sell every year plan to introduce things like lane sensors and radar gradually.
The reason for that decision has grown to legendary proportions within the industry, and Krafcik displayed slides of the worst of it: Google’s own employees putting on makeup and texting while behind the wheel of their then-semiautonomous test cars. The last slide showed a driver sleeping at 55 mph.
“We shut down this aspect of the project a couple of days after seeing that,” he said.
And Waymo and its competitors also will have to persuade a semi-skeptical public about the safety of a radically different mode of transportation. People in the business are fond of citing a particular number, and Krafcik did so too: 94 percent of crashes are caused by human error. That means 6 percent are caused by something else, and driverless carmakers know the glare of publicity will be excruciating when one of their cars, even if it’s the other car’s fault, gets into a crash.
Convincing the passenger that all is well is one reason Waymo has installed a pair of video screen monitors. In the case of the Pacifica, they are recessed into the [otherwise empty] front seats.
“We’ve obsessed over the right kind of information to show at the right time,” said Ryan Powell, who heads Waymo’s user experience design team. The amount of data collected by 360-degree sensors and cameras is immense, so Powell’s team is selective: other cars are blue blocks, pedestrians are white circles, bicyclists are a lighter blue circle and traffic cones are yellow. Crosswalks are lighter when someone is walking or riding in or near them, slightly darker if they are empty. Traffic lights? Red, yellow or green, of course.
“On one hand, we’re really excited to show off just how much our car can see,” Powell said. “At the same time, we don’t want to overwhelm users.”
Though some pavement hails from the Air Force base, which closed in 1995, Waymo has created a maze of roads, traffic signals, curbs, driveways, potholes, a water trough and a working railroad crossing to test its cars.
“Structured tests allow us to model and stage challenging scenarios that exist in the real world,” said Stephanie Villegas, a systems engineer who heads the structured testing team.
For media day, they put on three of them. In one, a Honda convertible abruptly cut off one of the Pacificas just past an intersection, and the Pacifica braked to avoid it. In another, a car obscured by two parked vehicles suddenly backed out of a driveway. Again, the Pacifica came to a halt. The third was a faux moving-day scene, with a couch in the street and boxes tumbling into the lane, just as another car approached from the opposite direction. The Pacifica knew better than to swerve into the car. It braked until the coast was clear.
After an estimated 20,000 such tests since 2012, Villegas seemed delighted to have an audience to show off to.
“It’s so fun!” she said. “I’m so glad you’re all here.”