Silicon Valley Stumbles in World Beyond Software
Google parent’s struggle to launch a delivery-drone business is part of tech’s broader problem with physics; no space elevators or jetpacks
By Jack Nicas
Google then scrapped the drone and started over.
One problem was how the wind often toppled the device on takeoff or landing. “It was a dumb thing about physics,” says Chris Anderson, chief executive of 3D Robotics Inc., which has made parts and software in Google’s drones.
Google parent Alphabet Inc. and others in Silicon Valley are broadening their sights from the digital to the physical world in a bid to expand their influence, and their bottom lines. They promise to reinvent everything from cars to thermostats to contact lenses. Yet in a sign of how innovation is stalling broadly in the American economy, they are finding their new terrain far harder to control than their familiar digital turf.
Alphabet’s 58 self-driving cars have traveled 2.2 million miles, but they are still flummoxed by snow and drive so conservatively they can disrupt traffic. Its high-altitude balloons designed to beam internet to remote areas have sometimes crashed in shreds, baffling engineers. A planned interactive jacket was delayed for a year in part because its sensor-embedded threads snapped under the pull of industrial looms. The tech giant abandoned projects involving cargo blimps, vertical farming and seawater-to-fuel technology that proved too difficult or expensive.
Others face similar problems. Facebook Inc. is struggling to launch solar-powered drones to beam internet connections via lasers. Starship Technologies, led by a co-founder of video-chat service Skype, plans a fleet of delivery robots, but current models use human operators to cross a road.
In software, programmers can control their environment. The physical world is messy and unpredictable. Even the smartest computers can’t prepare for every possibility. Add to that the burden of public safety and regulation and it is easy to see why the tech industry hasn’t been able to replicate its success in the digital realm.
“The world is so unforgiving. You can’t just ask it to be more organized,” said Astro Teller, the ponytailed chief of X, Alphabet’s research lab that has investigated—and decided against—space elevators and jetpacks.
Moreover, digital progress is rapid, because computing power increases dramatically over time and software can be replicated endlessly. In the physical world, advances are constrained by physics.
Silicon Valley’s push into the physical world “is going to be a much longer, slower process, especially in the next couple of chapters,” said Andrew McAfee, a Massachusetts Institute of Technology professor who studies technology’s impact on society.
That timeline isn’t typical for Google. Larry Page and Sergey Brin launched the search engine—then called BackRub—in 1996 on a single computer in a Stanford University dorm room. By 2000, the service handled 60 million queries a day.
That year, Google director and early investor John Doerr said, “The new economy is based on bits, not atoms,” a reference to the ones and zeros of computer code.
These days, Alphabet is very busy with atoms—in cars, robots, drones, smoke detectors and wind turbines, to name a few. Current and former employees, say Messrs. Page and Brin tell workers to focus on “atoms, not just bits.” Mr. Doerr said in an interview that he believes the founders are still more focused on bits.
With that mind-set, Google in 2010 formed X, its factory for “moonshots,” or projects that tackle a big problem that could take as long as a decade. Its roster includes Project Wing, the team of roughly 85 workers designing drones to ferry packages. The Wall Street Journal compiled an account of the team’s efforts via interviews with current and former employees.
To Alphabet executives, drone deliveries are the next stage of logistics—the 21st-century equivalent of the Pony Express. Google formed its delivery-drone team in 2012 and hired Massachusetts Institute of Technology robotics professor Nick Roy to run it. He brought along two graduate students who had built a winged drone that could fly autonomously through an MIT parking garage.
Unbeknown to Google, Amazon.com Inc. was already developing its own delivery drones, which it disclosed in 2013.
At first, X considered using drones to ferry defibrillators to heart-attack victims, but abandoned the notion after realizing that approach is no better than an ambulance. Says one former employee, “When it doesn’t work, is [the news headline] going to be: Google Drone Kills Elderly Man By Being Late?”
Project Wing shifted to drones to deliver small parcels—from food to electronics to medicine—and bolster Google’s role in logistics, where it was starting a shopping-delivery service.
The few civilian drones available in 2012 were so-called quadcopters, which take off vertically and hover. But copters are inefficient; their batteries typically can power only 20 minutes of flight, less when carrying a package. The alternative was a winged drone, which can fly farther but can’t hover.
It also needs a runway to take off and land.
Mr. Roy said he wanted a drone that could do both—take off like a copter and glide like a plane. They settled on a so-called tailsitter—a glider that sits upright and uses propellers to take off vertically, then rotates in air to glide horizontally.
Engineers cycled through a series of prototypes—code-named the Auk, the Bauk and the Super Bauk—before settling on a white-foam-and-plastic glider dubbed the Chickadee that was roughly 3½ feet long and weighed 18 pounds.
Each version brought problems. Flights that worked in software simulations didn’t work in the real world. The drone didn’t follow flight paths precisely. The device toppled when trying to land on its tail. Antennae became disoriented when the craft rotated in midair.
Mr. Anderson, who has provided parts for Google drones, said the drone suffered from “the spork problem, where you try to get something to do two things and it does neither very well.”
Testing software is straightforward. Changes can be simulated at the push of a button, and flaws can be spotted and fixed. Google is famous for testing software extensively. It constantly tweaks its search algorithms to make sure they are delivering the most relevant results for users.
Beyond designing an aircraft, the Google team faced an equally vexing challenge: how to deliver a package.
Engineers shadowed deliverymen from Google Express, the delivery service. They saw the wide range of delivery locations, from homes with back entrances to apartment complexes with courtyards. Shippers often struggle with the “last-mile problem;” one former X employee called it “the last-inch problem.”
They struggled to devise a solution. “Do you land? Do you lower it on a string? If you land, are people going to steal the drone? Are they going to be afraid of it? How dangerous is it going to be?” the former employee said. “There was endless debate—I mean months—on landing versus not landing.”
The team eventually settled on lowering the package on a winch. But that required the tailsitter to hover upright, creating a sort of sail in the wind.
For deliveries, both Alphabet and Amazon now are considering asking customers for help. In a scenario depicted in an Amazon video and that former X employees said Alphabet also is considering, customers would place a specially marked mat as a landing pad that a drone could recognize from above.
One former X employee compared the notion to “having to go out and meet the mailman when he gets there.”
By early 2014, Google co-founder Mr. Brin, tiring of the delays, set a deadline: Make a delivery to a non-Googler in five months. That led to the dog-treat drop in Australia.
After that, “we threw everything out,” one former employee said. “Everything. I mean not just the form factor…[Everything] was deemed bullshit.”
Mr. Roy returned to MIT. Mr. Teller replaced him with Dave Vos, a South African aerospace engineer.
The bald, cerebral Mr. Vos brought a rigor unfamiliar on the famously loose Google campus, aiming to replicate the strict rules of the aviation industry. He ordered frequent tests of drone designs, including in a wind tunnel. In early 2016, Mr. Brin moved his desk into the Project Wing offices to help push the team along.
Wing’s new drone resembles a catamaran, with a 5-foot wing laid across two thin poles that each support two rotors and a fin. Like the tailsitter, the device takes off vertically and glides horizontally, but doesn’t require midair rotation. X has registered 75 drones dubbed “hummingbird” with the FAA under the alias Ashfloyd LLC, according to registration numbers shown in an X video. Amazon is testing a similar design.
Like the tailsitter, the new drone is unreliable, former employees said. Over months of tests earlier this year at Flying M Ranch, a 20-square-mile cattle ranch in Merced, Calif., X’s drones repeatedly crashed, wandered off, lost power or tried to land in trees. One former employee said the goal was to complete 1,000 flights without an incident but they could never pass 300.
X aims to create an online exchange dubbed Wing Marketplace where customers can get orders from retailers and restaurants delivered within minutes via drone, former employees said. X would charge customers a $6 drone-delivery fee. X has met with Whole Foods Market Inc., Domino’s Pizza Inc. and a series of other restaurant and fast-food chains, former employees said. X’s control over the user experience killed talks with Starbucks Co. to deliver coffee, the employees said. Whole Foods, Domino’s and Starbucks declined to comment.
In September, Project Wing completed its second deliveries to non-Alphabet employees. Its drones ferried burritos over a closed field from a Chipotle Mexican Grill Inc. food truck to some Virginia Tech staff and students a few hundred feet away.
A few weeks later, Mr. Vos and his top commercial executive resigned. Alphabet executives pushed out the pair over conflict with the team’s top engineers, according to people familiar with the situation. Talks for a test in Ireland are now on hold.
Mr. Teller says obstacles and setbacks are the nature of X’s work to innovate in the physical world.
“The moonshot factory is a messy place. But rather than avoid the mess, pretend it’s not there, we’ve tried to make that our strength. We spend most of our time breaking things,” he said in a TED Talk earlier this year.
A spokeswoman says X remains “wholeheartedly committed” to delivery drones and isn’t downsizing Project Wing. “We believe that opening the skies to faster, more efficient transportation of goods is a moonshot worth pursuing,” she says. X aims to carry out more tests next year.
Amazon, meanwhile, is testing its delivery drones in the U.S. and Cambridge, U.K., while a startup named Flirtey Inc. last month started delivering Domino’s pizza via drone in a small test in New Zealand.
One major challenge to tech’s push into the physical world is its high stakes. For the first time, tech firms are crafting devices that could hurt people. Last year, the propeller of a hobbyist’s runaway drone sliced a U.K. toddler’s eyeball. The potential dangers mean that regulations could hamstring tech companies accustomed to launching products on their own schedules.
But former employees say regulations aren’t the main problem. One former X worker says Alphabet is “a software company—not an airplane company.”