Waymo’s driverless car: ghost-riding in the back seat of a robot taxi
I’m in the middle seat of a Chrysler Pacifica minivan, heading north on Dobson Road in Chandler, Arizona, when I notice we may have taken a wrong turn. Under normal circumstances, I would just lean forward and ask the driver for an explanation. But in this case, that’s not possible. There is, after all, no driver to ask.
Last October, Alphabet’s self-driving subsidiary Waymo emailed its customers in the suburbs of Phoenix to let them know that “completely driverless Waymo cars are on the way.” For several years, Waymo has offered its autonomous taxi service to a small group of people, but the rides typically included a safety driver behind the steering wheel. Now, Waymo is saying more of those rides will take place sans safety driver, a sign that the company is growing confident in the accuracy of its technology.
Weeks later, I’m in one of the company’s driverless minivans going the wrong way. But before I can ponder my new life as a victim of robot kidnapping, the vehicle realizes its mistake and reroutes. Rather than taking a left on West Baseline Road, we ended up going right. (A spokesperson later confirmed that the vehicle determined it couldn’t get over in time to make the left, and instantly rerouted.) A quick detour through a residential neighborhood puts us back on course. It’s a minor error, but I can’t help but wonder what other surprises await me during my brief trip in one of the world’s only fully driverless vehicles operating on public roads today.
It’s not a big deal, but I’ve ridden in a fully driverless Waymo once before. It was October 2017, but that ride was on a private road — actually a decommissioned Air Force base in California’s Central Valley, which is now owned by Alphabet — completely separated from the chaos and unpredictability of US roads. I’ve taken a total of four trips with Waymo, mostly with trained safety drivers in the front seat. All of them have been mostly uneventful. Honestly, I’m at the point in my career where riding in a self-driving car just doesn’t do it for me anymore. The thrill is gone.
But this is different. The steering wheel spins on its own. The pedals go up and down as if pressed by some phantom foot. The seatbelt is clicked into place but technically protects nothing. The vehicle drives more assertively than it has in previous trips, breezing past construction sites at 45 mph with barely a twitch and nudging its way into intersections in a way that feels less robotic than it has in the past. The construction site, with its orange cones and hard-hatted workers milling around, is particularly impressive.
I’m trying to keep my composure, but it’s oddly thrilling. I keep leaning forward to look at the empty seat to assure myself this is really happening.
It is also entirely mundane, as any 15-minute ride-hailing trip through suburban America in 2019 would be. The future of transportation is here, and it’s driving the wrong way past row after row of soulless terra cotta strip malls. I keep glancing over at other drivers on the road, expecting looks of shock and amazement. Check me out in this driverless car!
No one looks.
There are, of course, many caveats to consider.
Waymo has been testing its vehicles in the Phoenix area since early 2017. Its self-driving cars operate in an approximately 100-square-mile service area that includes the towns Chandler, Gilbert, Mesa, and Tempe. In late 2018, the company launched a limited public ride-hailing service called Waymo One, but the only customers to get access were people who had first been vetted through Waymo’s early rider program of beta testers. Last week, Waymo said it has around 1,500 monthly active users from both programs.
“It’s been an enormously difficult, complicated slog, and it’s far more complicated and involved than we thought it would be,” says Nathaniel Fairfield, who leads the team that oversees the decision-making part of Waymo’s onboard software. “But it is a huge deal.”
The geofence, or the defined geographic area, for the company’s fully driverless vehicles is much smaller, about half the size of the overall service area — or about 50 square miles. The vehicles technically can’t pick up or drop off passengers outside that zone. Even more complicated: Waymo’s fleet operations center is located outside the geofence, so a Waymo employee needs to chaperone the fully driverless-enabled vehicles into the zone each time before it makes pickups.
Only members of Waymo’s early rider program are allowed to ride in the company’s driverless vehicles. Those people sign nondisclosure agreements with the company in order to get access to early versions of Waymo’s technology. This bars them from speaking publicly when, say, one of their trips goes off course. (A review of customer feedback from 10,500 trips earlier this year by The Information found 70 percent of Waymo’s trips received the highest rating of five stars, while 30 percent were rated four stars or less.) Waymo won’t say how many people or how many trips its fully driverless vehicles have made so far.
Oh, and “fully driverless” isn’t the only way the company describes these trips. Waymo also prefers “rider only.” My guess is a focus group told them that sounded less scary than “fully driverless.”
There are other conditions to the driverless rides. I was surprised to learn that Waymo lets them drive at night, but not when it’s raining or during Phoenix’s frequent dust storms (which are known to locals as “haboobs”).
These driverless cars aren’t totally alone in the wilderness. Waymo has a team of remote employees that watch the real-time feeds of each vehicle’s eight cameras and can help, with the push of a button, if the software runs into a difficult spot and needs a human eye to figure out what’s going on. I counted at least two cameras in the headliner watching me, but no one from the rider assistance team checked in during the ride.
“These folks don’t joystick the car or anything like that,” Fairfield says. “But they can help answer specific questions that a car might have about an ambiguous situation and that’s where human intuition and human understanding of the entire context is super important.”
For example, if the Waymo vehicle came across a moving van with the doors open and people loading and unloading things, it wouldn’t be able to determine if the vehicle was getting ready to move or staying there for a while. “That’s not something we’ve gotten around to making the car smart enough to understand,” Fairfield says. “But a human sees it in a moment and can send that signal, [though] it’s not really a command to the car.”
Waymo avoids making projections about when driverless cars will go mainstream, but most experts agree that early self-driving predictions were overly optimistic. A starry-eyed assessment is 10 years. Many others say decades as researchers try to conquer a number of obstacles. The vehicles themselves will debut in limited, well-mapped areas within cities and spread outward.
This isn’t the first time Waymo has unveiled fully driverless cars to much fanfare. On November 7th, 2017, Waymo CEO John Krafcik took the stage at a tech conference in Lisbon, Portugal, and said, “Fully self-driving cars are here.” He showed a video of the company’s driverless vehicles picking up and dropping off passengers in Chandler. And he promised that “within the next few months,” more people would get the chance to ride in Waymo’s most advanced vehicles.
But that didn’t really happen. The company released another video in May 2018 that was pretty similar to the first: passengers giggling nervously at the sight of an empty driver’s seat, wondering aloud whether passersby are also slightly freaked out, and making casual references to “the future.” One man gets so bored that he falls asleep.
Those trips occurred in a small, largely residential portion of its service area, and they were never a majority of the rides provided by Waymo. Meanwhile, reports begin to circulate about problems with Waymo’s driving. The company’s most advanced vehicles were still occasionally confounded by certain traffic situations, which suggests the tech — while incredibly advanced — was not quite ready for the public rollout Waymo was predicting. By mid-2018, the company began putting trained safety drivers back in the driverless vehicles, according to The Information. And the company stopped talking publicly about its driverless trips.
I’m standing in the middle of Waymo’s 39,000-square-foot operations center, surrounded by rows of gleaming white (or, in some cases, not-so-gleaming white) minivans. It’s the mid-afternoon shift change, and dozens of the company’s safety drivers are bringing their self-driving cars back to the depot to get cleaned, refueled, and recalibrated before getting sent back out into the Arizona haze to look for more fares.
Waymo plans to open another op center in Mesa, a sign the company plans on scaling up in the Phoenix area. Another sign: outside the Chandler facility, two of the company’s autonomous 18-wheeled tractor-trailers are parked, getting prepped for another round of testing.
I learn that the cone-shaped roof-mounted LIDAR sensors — arguably the most important piece of hardware on the self-driving car — can get really filthy during each shift. Enough dirt builds up, and the LIDAR’s ability to send out hundreds of thousands of laser beams to map the surrounding environment can be diminished. But with the touch of a button, tiny windshield wipers pop out, then twirl around the surface of the sensor amid a spray of cleaning fluid. Presto: clean LIDAR.
Off to the side sits a tiny, egg-shaped vehicle with a dusty LIDAR cone of its own. This is the “Firefly,” Google’s driverless car that famously gave a ride to Steve Mahon, a man with severe visual impairments, through Austin, Texas, in 2016. This is widely considered to be the first fully driverless car test on public roads. The steering wheel-less, pedal-less prototype was retired in 2017, but some op center employees still use it as a phone booth (albeit a non-soundproof one).
In a nearby break room, safety drivers lounge on couches, scarf Kind bars, and watch TV. A poster on the wall features a man passed out at the steering wheel, with the text reading, “Say No to Fatigued Driving.”
Fairfield doesn’t seem fatigued — far from it. He exudes raw, dad optimism about the company’s second, and by his telling, more confident pivot to fully driverless trips. Still, he acknowledges that Waymo may have underestimated the sheer complexity of the task it set out to do.
“The world is a very complicated place,” Fairfield says. “This is a much bigger problem than we sort of imagined it 10 years ago when we got started. You know, we had a sort of a simplified view, a layman’s view of driving. Now we kind of understand all of the nuance and complexity that goes into it.”
The driverless cars are only allowed to do pickups and dropoffs within the 50-square-mile geofence, but Waymo is already testing its vehicles beyond those boundaries in the hopes of broadening the service area.
In a minor victory for my reputation as a journalist, I’m able to convince Waymo to let me choose the pickup and dropoff locations. Many experts believe this is an important test for a truly driverless ride-hailing service: the ability to route dynamically, without any pre-planning, to a destination. They ask to vet the locations first to make sure they conform to the vehicle’s limited geofence, and I agree because I am beneficent. The negotiation takes several e-mails.
There’s a lot of skepticism that driverless cars will be the traffic safety panacea that Waymo and other technologists claim. In March 2018, a 49-year-old woman was struck and killed by a self-driving Uber vehicle while crossing the street in Tempe, Arizona. A panel of federal investigators later determined that the cause of the crash was a distracted safety driver.
Waymo has long argued that driver-assistance technology like Tesla’s Autopilot will just exacerbate the problem. Only by completely eliminating human involvement can we be assured that driverless cars are safe, they argue. But is Waymo getting around this problem by removing the driver, or actually creating new complications altogether?
Fairfield offers his assurances. “You can tell yourself that you’re safe as much as you want,” Fairfield says. “But when it really comes down to brass tacks and you’re taking that driver out, it makes it real in a way, and the entire team takes it very, very seriously. Everybody looks in every closet, under every rock, under every rug, to think of any issue that they might have. And you flush all that stuff out and you look at it and you solve it and you work on it. And then you realize that you’re ready to go truly driverless.”
The driverless car has one more surprise in store for me. Toward the end of the trip, as we’re moving slowly through the parking lot of a restaurant called the Watershed, the minivan comes to an abrupt halt: a flock of pigeons is in the way.
Is this a false positive? A sign that autonomous vehicles will be braking for every last animal or floating plastic bag that enters its field of view? I have no great love for pigeons, but I support the car’s caution. All creatures great and small, after all.
Later, I’m in an Uber on the way to the airport. I have a great conversation with the driver during which I learn all sorts of fun facts about living in Phoenix (like the aforementioned haboobs). This gets me thinking about what we lose when we go driverless. Not every Uber driver is a worthy conversationalist, but I think we can all agree that IRL human interaction is a dying breed. I understand this is not a universally held opinion.
But this is a trivial concern. The implications of fully driverless cars are mind-boggling, regardless of whether you believe they’ll ever truly go mainstream or remain bound by their tiny geofences in cities around the globe. The optimists envision rides so cheap that personal car ownership becomes a thing of the past. No more distracted driving. No more fatalities. Unpave the parking lots, and put up a paradise.
The skeptics wring their hands over the possibility of robot gridlock, of public transportation left to mothball and crumble, of self-driving cars priced at a premium so only the wealthiest can afford their promises of safety, of algorithms written in a way to play god.
Whatever the outcome, the technology isn’t theoretical anymore. Driverless cars are here. There’s no going back now.