Inside Google's Secret Drone-Delivery Program

For two years, the company has been working to build flying robots that can deliver products across a city in a minute or two. An Atlantic exclusive.

A zipping comes across the sky.

A man named Neil Parfitt is standing in a field on a cattle ranch outside Warwick, Australia. A white vehicle appears above the trees, a tiny plane a bit bigger than a seagull. It glides towards Parfitt, pitches upwards to a vertical position, and hovers near him, a couple hundred feet in the air. From its belly, a package comes tumbling downward, connected by a thin line to the vehicle itself. Right before the delivery hits the ground, it slows, hitting the earth with a tap. The delivery slows, almost imperceptibly, just before it hits the ground, hardly kicking up any dust. A small rectangular module on the end of the line detaches the payload, and ascends back up the vehicle, locking into place beneath the nose. As the wing returns to flying posture and zips back to its launch point half a mile away, Parfitt walks over to the package, opens it up, and extracts some treats for his dogs.

The Australian test flight and 30 others like it conducted in mid-August are the culmination of the first phase of Project Wing, a secret drone program that’s been running for two years at Google X, the company’s whoa-inducing, long-range research lab.

Though a couple of rumors have escaped the Googleplex—because of course Google must have a drone-delivery program—Project Wing’s official existence and substance were revealed today. I’ve spent the past week talking to Googlers who worked on the project, reviewing video of the flights, and interviewing other people convinced delivery by drone will work.

Taken with the company’s other robotics investments, Google’s corporate posture has become even more ambitious. Google doesn’t just want to organize all the world’s information. Google wants to organize all the world.

During this initial phase of development, Google landed on an unusual design called a tail sitter, a hybrid of a plane and a helicopter that takes off vertically, then rotates to a horizontal position for flying around. For delivery, it hovers and winches packages down to the ground. At the end of the tether, there’s a little bundle of electronics they call the “egg,” which detects that the package has hit the ground, detaches from the delivery, and is pulled back up into the body of the vehicle.

The Google delivery drone releasing a package (Google)

That Parfitt would be the man on the receiving end of the tests was mostly happenstance. Google’s partner in the country, Phil Swinsburg of Unmanned Systems Australia, convinced him to take part in the demonstration deliveries launched from a nearby farm. (Australia’s “remotely piloted aircraft” policies are more permissive than those in the United States.)

Standing with Parfitt as he received dog treats from a flying robot was Nick Roy, the MIT roboticist who took a two-year sabbatical to lead Project Wing. In all the testing, Roy had never seen one of his drones deliver a package. He was always at the takeoff point, watching debugging information scroll up the screen, and anxiously waiting to see what would happen. “Sergey [Brin] has been bugging me, asking, ‘What is it like? Is it actually a nice experience to get this?’ and I’m like, ‘Dude, I don’t know. I’m looking at the screen,’” Roy told me.

So, this time, as he prepared to end his tour of duty at Google X and return to MIT, he watches as the Wing swoops and delivers. Recalling that moment, he struggles not to sound too rapturous or lose his cool technical objectivity. “Once the package is down and the egg is back up, the vehicle gains altitude, and does this beautiful arc, and it’s off again,” he said. “That was delightful.”

Google

The parting between Roy and Google X seems amicable. When Astro Teller, director of the lab, described it to me in an interview in Mountain View, he literally patted Roy on the knee. “Nick was super ultra-clear with us from day one, despite lots of pressure from me,”—Teller pat Roy on the knee—“that he was going to leave after two years.” But the timeline was good, Teller maintained, because it gave the project shape and a direction.

In the two years, Roy’s goal was simple: figure out if the idea of drone delivery made sense to work on. Should Google pursue creating a real, reliable service? Was it possible? Could a self-flying vehicle be built and programmed so that it could take off and land anywhere, go really fast, and accurately drop a package from the air?

The answer, Roy and Teller say, is yes. They have not built a reliable system Google users can order from yet, but they believe the challenges are surmountable. Now, Google will begin growing the program in an ultimate push to create a service that will deliver things people want quickly via small, fast “self-flying vehicles,” as they like to call them.

Teller has found a replacement for Roy in Dave Vos, a 20-year veteran of automating flying machines, who sold his drone software company, Athena Technologies, to Rockwell in 2008. Where Roy got to play what-if and why-not, Vos must transform the Wing into a service that real people might use.

“What excited us from the beginning was that if the right thing could find anybody just in the moment that they need it, the world might be radically better place,” Teller said.

There are already dozens of Googlers working on the project, concocting everything from new forms of the vehicle to the nature of its delivery mechanism to the user experience of the app for ordering drones. There will be more recruits soon. Google will enter the public debate about the use of civilian unmanned aerial vehicles. Regulators will start hearing from the company. Many packages will be dropped from the sky on a tiny winch from a robot hovering in the air.

This may sound crazy. This may be crazy. But Google is getting serious about sending packages flying through the air on tiny drones. And this is how that happened.

* * *

Of course Google wants the world to believe in delivery by drone as part of the natural progression of technological society to deliver things faster and faster. This is how the world works, according to Google co-founder, Sergey Brin.

Imagine Brin in 2011. Perhaps he’s wearing a Google Glass prototype and a long-sleeved technical t-shirt, maybe even Vibram FiveFinger footwear. He is rich beyond all comprehension, a billionaire many times over. In his 39th year on Earth, he has decided to grow a beard, wisdom-enhancing salt-and-pepper sprinkled around his chin.

While Larry Page runs the mainline cash cow Internet advertising business, Brin (or Sergey, as everyone at Google X invokes him) is building a second, much wilder company inside the envelope of the old one. Over the next few years, he will unveil self-driving cars, Google Glass, help acquire eight robotics companies and a high-altitude, solar-powered drone maker, and do whatever else Google is doing in secret.

And one day in 2011—before any of us had seen these new ideas—he is talking with Astro Teller, whose goatee is more salt than pepper, and they make an observation about the world.

“The original observation felt most like this,” Teller said. “When the Pony Express came along, it really reshaped society to be able to move things around fairly reliably at that speed, which was measured in many days. The U.S. Postal Service—growing partly out of the Pony Express and having it be even more reliable and starting to shorten the time—really did change society again.

Library of Congress

“FedEx overnight delivery has absolutely changed the world again. We’re starting to see same-day service actually change the world,” he continued. “Why would we think that the next 10x—being able to get something in just a minute or two—wouldn’t change the world?”

If there is one thing Google likes, it is changing the world. The company’s framework for societal transformation has been conditioned by the relentless decrease in cost and increase in performance of computers. They believe order-of-magnitude changes can happen quickly because they’ve seen and participated in both the rise of the commercial web and the astonishing growth of mobile computing.

To these technical changes, they attach the concept of progress, especially if Google, with its deeply held sense that it won’t or can’t be evil, is involved. As the company has matured, people like Teller seem willing to admit that perhaps all things aren’t getting better all the time. But they argue the new “goods” outweigh the new “bads,” especially if an honest accounting is made of the current alternatives.

“Google X has this experience all of the time in all of these different projects,” Teller said. People count all the problems created by our current way of life as zero because that's what we’re used to as the societal default, he contended. Conversely, people immediately see the negatives of any new thing. “We are not deaf to those issues and we’re really eager to talk to society about how to mitigate those,” Teller said. “But part of our conversation with society is about us listening, but also trying to remind the people that we talk to that the place we’re starting from is not zero. In this case, for delivery, cars, airplanes create a very large carbon footprint and have a lot of safety issues.”

So, of course Google wants to help increase the speed of delivery and reduce the carbon footprint and safety of delivery. Ergo, the development of self-flying vehicles. “In principle that [speed improvement] could happen independent of self-flying vehicles,” Teller concluded. “But it was obvious from the very beginning that it was going to have to be self-flying vehicles.”

Google X began to come up with ideas and test them theoretically and experimentally. They considered many different wild options, sketching out new and wacky transportation systems. (“What if you took a glider up on a balloon with a super long string and the glider goes up, releases, and zooms down… You can—on paper—satisfy yourself that’s not the right solution.”) But eventually, Teller realized they needed an expert. They did a search and ended up pulling Roy across the county.

Roy was perhaps a less-than-obvious choice. For one, he’d never worked on drones flying outside. The challenges of the wind were new to him. Roy neither had a traditional aeronautics background nor had he dealt in logistics. Look back on his resume from the early 2000s, as he prepared to finish his PhD at Carnegie Mellon: There are almost no signs that he’d be the guy Google X would one day tap for a drone project. His most prominent work had been on tour guide and nursing robots.

Nick Roy in Australia during the Wing delivery tests (Google).

But that leaves out one very important detail: Roy's thesis advisor was Sebastian Thrun, the founder of Google X, and one of the most influential people in robotics. In the years before his tour at Google, Roy did important work with the support of the Office of Naval Research on indoor drone navigation in "GPS-denied" environments, where the vehicles can't rely on satellites to position themselves.

When Roy arrived in California, Project Wing’s initial focus was on delivering defibrillators to help people who have had heart attacks. The key factor in the success of using a defibrillator is how quickly it is deployed, so saving a few minutes of transit time could make for a lifesaving application. But as time went on, the Google team realized that tying into the 911 system and other practical exigencies eliminated the speed advantage they thought they could deliver.

So, now, Teller’s—and, by extension, I will assume Brin’s—big-picture vision has shifted to the ways ubiquitous, two-minute delivery can transform people’s relationship to stuff.

The idea goes like this: Because people can’t assume near-instantaneous delivery of whatever they need, they stockpile things. They might have a bunch of batteries, slowly decharging in a drawer, or a drill that they use for 10 minutes a year. Each of these things is a personal possession that sits around, embodying all this energy and industrial effort unproductively.

If this sounds familiar, it should: It is the argument—even down to the drill example—that organizations like Worldchanging made in the mid-00s for the creation of “product-service systems.” Those ideas, in turn, became key planks in the original conception of the “sharing economy,” imagined as one in which the world could make much less stuff because efficient, digital logistics would let each asset be used by more people.

“It would help move us from an ownership society to an access society. We would have more of a community feel to the things in our lives,” Teller preached. “And what if we could do that and lower the noise pollution and lower the carbon footprint, while we improve the safety of having these things come to you?”

And unicorns might win the Kentucky Derby, too! But one would need to find a unicorn first before it could enter the race.

Google had to build a vehicle and teach it to fly itself.

* * *

The home of Project Wing (Alexis Madrigal)

Off one of the many hallways inside Google X’s simple red brick building in Mountain View, there is a door labeled "The Hatchery." Roy swipes his badge and we step inside the guts of the secret Project Wing.

This is a workshop. Scattered about, I can see fishing line on a table, three colors of tape, a tall half-stuffed trash can, drawers of fasteners, spare antennae, several glue guns, and some drills. Off to my right, through glass doors, there are four identical plane bodies lined up, wingless. At the back, a man is hand-building some electronics, copper gleaming in the overhead lights.

Carapaces of different species of unmanned aerial vehicles are piled on shelving units and anywhere else they might fit. The horseshoe-crab shaped bodies of several editions of the current drone sit down at my feet. Their electronic innards are visible through clear plastic. Above me, a Cessna model hangs from the ceiling. On shelving units, there are the familiar bug-like quadcopters, a strange craft with helicopter rotors built into its single wing, and a remote control monster truck.

The main attraction, however, is the gleaming white prototype sitting atop the wheeled table in the center of the room labeled Chickadee. It sits on its tail in the angle of repose of the Space Shuttle, nose pointed to the sky. This is the tail-sitter, just like the one that dropped the dog treats in Australia.

The design is simple. There is a tail that serves as a stand, a central plastic body, and two wings made out of foam board covered in a thin skin for protection from the elements. There are four rotors attached to the vehicle, two on the underside closer to the body, and two on the outside towards the edges of the wings.

The build quality is fascinating. From afar, it looks shiny and complete, and it’s loaded with custom-built electronics, but up close, it’s clear that the body itself is handmade and hacked together. Fingerprint smudges smear it. Some pieces have been professionally fabricated, it seems, but other bits look made in the on-site shop. It is a work in progress.

The class of vehicles that it belongs to is not common. Most flying things are fixed wing—like a plane—or some type of helicopter, which uses one or more rotors to stay in the air. To fly, fixed wing craft primarily move air horizontally, while helicopters move air vertically. The tradeoffs are pretty obvious: The fixed wing craft are more aerodynamic and efficient. They can go farther, faster with less fuel. Meanwhile, the choppers can maneuver well in many different conditions, don’t need a runway to take off or land, and can hover in place.

In the military drone world of Predators and Global Hawks, fixed wing, long-range craft predominate. In the hobbyist drone world, quadcopters like the DJI Phantom 2 and Parrot AR.Drone are most popular among enthusiasts, although a strong model-airplane community exists.

In aeronautics, hybrid craft that combine elements of fixed-wing planes and helicopters do exist, and certainly aerospace companies experimented with them. But they are more complex because they have to execute two entirely different tasks: moving air on different axes. In some cases, such as the new F-35B, Lockheed Martin built rotating jets into the plane body that can be pointed at the ground to achieve liftoff, then rotated in the air, to push the jets through the sky.

The tail-sitter configuration, in which the whole craft rotates from a vertical to a horizontal position, has also been a source of fascination through aeronautical history. The Nazis, for example, were considering such a craft. And there was an American defense program that resulted in the creation of two prototype aircraft by Lockheed and Convair. The photographs of these huge planes sitting vertically on runways—shiny and steel, unmistakably mid-century—feel retrofuturistic. None of the research efforts caught on, though, with a major problem being that there wasn’t a good way for the pilot to deal with the change in orientation.

The Lockheed XFV-1 tail sitter (US Navy).

Obviously, that’s not a problem with a drone, though. The “pilot” is housed in a desktop-class computer that sits towards the tail of the plane. The power system, batteries, cabling, and a big capacitor, sit just above it. That’s hooked to the motors, which also send back motor performance data to the flight computer. Sensor data also comes in from the inertial measurement unit (IMU) mounted to the left of the computer. The IMU uses accelerometers and gyroscopes to determine the X-Y-Z positioning of the craft, an essential part of flying. In the nose, we find the GPS unit, and in the tail, there’s a camera pointed down. There’s no on-board laser rangefinding system in the current incarnation, but there are two communications radios, one high-bandwidth for sending telemetry data, and one low-bandwidth for longer range communications.

Google has not settled on this design for all its future program development, but it has formed the platform for much of their testing. While the hardware is a significant part of the problem, they seem largely agnostic about which flying machine might ultimately serve their needs best. The real challenges, Teller and Roy insist, come in the design of the rest of the system like, for example, the delivery mechanism.

Imagine all the possible ways one might get something from high in the air down to the ground. How about a tiny parachute à la The Hunger Games? Roy’s team tried it. There was too much wind interference and they struggled with accuracy. How about literally firing them down, a ballistic approach? “We contemplated this,” Roy said. “And then Sergey walked out from under a balcony and we almost hit him in a drop test.” After that, they moved on.

Another obvious idea is to simply land the craft, drop the package, and then take off again. To test the premise, they brought in some of Google’s user experience researchers who queried people about how they might react to such a delivery.

What they found was that individuals could not be stopped from trying to reach for their packages, even if they were told that the rotors on the vehicle were dangerous, which they are.

Finally, they settled on an idea that Roy had initially resisted: winching down a line with the package on it and then winding it back up into the craft.

Mechanical engineer Joanna Cohen, trained at Cal Tech and MIT, designed the contraption. It consists of a few key parts. The first is the winch itself, which spools out the hi-grade fishing line. The second is the “egg,” the little gadget that goes down with the package, detects that it has reached the ground, releases the delivery, and signals that it should be cranked back up to the hovering UAV. If something goes wrong, there is an emergency release mechanism at the top of the line—“basically a razor blade,” Cohen told me—that allows the UAV to cut and fly.

When a package comes hurtling down, it moves at about 10 meters per second (about 22 miles per hour). When it gets close to the ground, the winch slows the fall to 2 meters per second for a relatively soft landing.

In the abstract, or under ideal conditions, this seems simple enough. But the project’s hardware lead James Burgess said that out in the world, it’s not so easy to make the deliveries work.

“If you can imagine a user case where we’re going to someone’s house, and the egg hits something—maybe it hit the power lines, maybe it hit the trees, maybe it hit the roof, maybe it hit the railing on the porch before it got to the porch. There are a lot of unknowns and environmental challenges,” Burgess said.

“So the egg is smart enough to know that it hit something, but the vehicle also knows how high it is and the winch also knows how much line it is letting out. The egg says, ‘I hit something,’ and the vehicle says, ‘But wait, you’re not far enough down yet, so keep going because probably you bounced off something and don’t arm yourself for [package] release.’ So, all of our sensors and components work together in this network to make good decisions.”

Or, for now, some kind of decision. When I asked how they planned to deal with power lines, which seem especially challenging to sense and avoid, the whole team demurred. “Remember: early days,” Roy intoned. “We’re not even close to that.”

* * *

Like all autonomous robots, delivery drones have three fundamental tasks. They have to understand their position in the physical world. They have to reason where they should go next. And they have to actually execute the control maneuvers to get there.

It turns out that the basics of getting from one place to another, under ideal conditions, are not that difficult. Some hobbyist drones can fly through a set of waypoints on their own. Others can follow a signal down on the ground. But these capabilities are more in the realm of autopilot than autonomy: They simply hold a bearing, altitude, and speed. It’s kind of like cruise control in the sky. It’s a pretty huge leap from cruise control to self-driving cars and the same is true of the jump from autopilot to self-flying vehicles.

But what is hard is dealing with the thousands of unexpected scenarios and “edge cases” that would inevitably crop up if these systems were deployed at scale. It’s the sum of how the vehicles handle all those difficult situations that add up to a reliable technology.

The analogy to Google’s self-driving car efforts is clear here: It’s not that hard to build software that can drive a car on the freeway or even around around Mountain View and deal with 99 percent of the things that happen.

But what about that one percent?

Finding and learning how to deal with all the possible edge cases, and coming up with safety procedures for what to do when the robot doesn’t know what to do is actually what forms the core of these big, long-term development programs.

Google's self-driving car software in early 2014 (Alexis Madrigal)

In self-driving cars, Google keeps a massive database of all the times when a human operator had to take control of a car. They can simulate what would have happened if the human had not tagged in, and try out different software approaches to teaching the system how to react, if, in fact, it would have made an error. Any time they change the system’s logic, Google tests the alterations against that whole database to make sure they haven’t broken something with the new fix.

Project Wing will probably adopt the same approach with both the database and the human operators. But instead of a single driver operating a single car, as has been the case in the autonomous vehicle program, Teller likes to imagine that there will be a relatively small number of operators controlling a number of drones, helping them make the right decisions in difficult situations.

“If a self-flying vehicle is trying to lower something and it goes down three feet and gets stuck, should it go home? Should it land? There’s not a right answer to that,” Teller told me. “That would be a good moment for it to raise its hand and say back to someone looking at the delivery control software, ‘What should I do?’”

This is a Google-y approach to the problem of ultra-reliability. Many of Google’s famously computation driven projects—like the creation of Google Maps—employed literally thousands of people to supervise and correct the automatic systems. It is one of Google’s open secrets that they deploy human intelligence as a catalyst. Instead of programming in that last little bit of reliability, the final 1 or 0.1 or 0.01 percent, they can deploy a bit of cheap human brainpower. And over time, the humans work themselves out of jobs by teaching the machines how to act. “When the human says, ‘Here’s the right thing to do,’ that becomes something we can bake into the system and that will happen slightly less often in the future,” Teller said.

One area where humans might be less helpful is the development of detect-and-avoid software that could help the drones deal with birds, other UAVs, helicopters, and the like. Some—some—of these issues could be solved by regulation that creates certain corridors or layers of air space for drones, as well as requiring transponders or other signaling mechanisms on all humanmade flying things. But that’s not a complete solution because as Teller put it, the birds aren’t going to wear instruments.

Roy says the project is still in the very early days of developing a mature, reliable detect-and-avoid system. But they are very far from having the right answers.

Think about what the problem really looks like: A camera or radar or laser is pointed at the sky in the direction that the vehicle is flying. The background could be either the sky or earthly terrain with all the variation that could imply. So the environment itself is pretty noisy. And the only signal that the drone was on a collision path with a distant object would be a few pixels in the image from, say, a camera. Working from that limited data, the software has to interpret those pixels as a type of flying thing and predict what it might do. And it has to do all that consistently under radically different lighting and visibility conditions.

Actual camera data from an experimental aerial object tracking system developed at ETH Zurich (Andreas Nussberger)

Predicting others’ flight paths requires that one’s algorithm make some tradeoffs. At one end of the spectrum, one could program the software to say that other flying things could do anything at any time. But that makes it incredibly difficult to fly in normal airspace and is overly conservative. On the other end of the spectrum, one could assign fixed and rigid paths to all other flying things, assuming they move more or less in straight lines along a trajectory. But that, too, could lead to problems if a plane turns or a bird dives or a quadcopter reverses direction.

In the self-driving car space, Google has also had to build these sorts of models for cars and pedestrians and bicyclists, but roads—and the logic of the roads—heavily constrain what maneuvers are likely. Furthermore, it’s easy to gather lots and lots of data about how drivers operate: All Google has to do is drive and drive and drive, loading ever more data into their models for how other vehicles move on the roads of California.

The sky is voluminous and these vehicles are small. It's a lot less crowded than the country's road networks, and flying things can move in all directions. Roy’s team found it difficult to even trigger their sense-and-avoid systems when they tried to do so intentionally by flying remote-controlled planes at them. So, the self-flying vehicles need these systems for ultimate reliability and autonomy, but they are exceptionally difficult to build—and to test.

There are other problems, too. The task of simply orienting the UAV in space can be difficult depending on GPS availability and accuracy. The cargo loading process requires lots of manual intervention. The economics of delivery might end up making no sense. The batteries need to get better. The vehicles need to get quieter. The reliability of the parts in the drones needs to go up.

Google also has to convince the public that they want drones instead of UPS trucks. This isn’t just about safety, but also the very real concerns that drone delivery might generate new kinds of airborne pollution, electronic locusts jittering across the sky. Or that it might destroy delivery truck driver jobs, which are some of the last good blue-collar gigs around.

And even more fundamentally: What the hell is anyone really going to use drone delivery of anything in two minutes service for? It’s a nice vision to consider the sharing economy delivered via robotic air, but what specific applications for these robots will actually make sense?

Recall that the initial application for drone delivery was sending defibrillators winging across cities. Well, many cities have solved this problem in a different way. They keep the machines geographically scattered across a city. That may be inelegant. That may be slightly wasteful. But it’s simple, it’s easy, and it does not require the invention and intervention of a flying robot.

* * *

Google, however, is not alone in thinking that delivery by drone is a plausible part of the future. Sure, there is Amazon, which announced a drone delivery development program last December. But there is also Andreas Raptopoulos and his company Matternet.

Forged out of some sessions at Singularity University, the off-the-wall futurology school in Silicon Valley, Matternet has been working to build a business around delivering medicines and other high-value goods in places without roads. They’ve tested in Haiti, the Dominican Republic, and Bhutan.

Since the Amazon announcement, interest in what they’re doing has exploded, and Raptopoulos expects a similar increase in attention with Google’s validation of their work. “We refer to our adoption curve as before- and after-Amazon. Things have really shifted in people’s minds. People have started thinking at the corporate and organization level. There is an opportunity to solve a big problem,” Raptopoulos told me. “And I think Google’s announcement with accelerate that even further.”

But Raptopoulous’ vision for the future of drone delivery is very different from Google’s. He imagines not an anywhere-to-anywhere free for all, but that drones will carry goods to landing depots run by local people who build their own small businesses around the UAV service. He doesn’t see this type of service cutting into the logistics business in rich countries, at least not for a long while.

There are other cargo drone believers, even outside Silicon Valley. In Europe, there is an entire organization—the Platform Unmanned Cargo Aircraft (PUCA)—devoted to bringing people together around the idea. Their vision of the future would see large cargo planes carrying between 2 and 20 tons of cargo flying relatively slowly and cheaply from places underserved by the existing infrastructure. One controller on the ground could handle 10 to 30 cargo planes flying at less than 300 miles per hour to save fuel. They could travel at all times of night and day, creating a more flexible in-filling logistics service to the current cargo system. In this scenario, cargo drones are like flying buses, not the speedy vanguard of two-minute delivery.

Founded by Dutch business school professor, Hans Heerkens, PUCA hosted a conference earlier this year that saw presentations from Airbus Defense & Space, the Dutch Air Force, and—most intriguingly—the journalist and novelist, Jonathan Ledgard, who is heading up a project with the Swiss Federal Institute of Technology around cargo drones for Africa.

Ledgard, who wrote one of the best novels published this decade in Submergence, shared a draft of their vision with me—and it is fascinating in its mix of high and low technology, pessimism and optimism. He calls the robots in his plan “donkeys.”

“The qualities of a donkey are similar to what is required for a cargo drone: surefooted, dependable, intelligent, able to deal with dust and heat, cheap, uncomplaining,” Ledgard wrote. “The choice of the name ‘donkey’ for cargo drones is deliberate. A donkey is not a Pegasus, associated with speed. It does not bomb, does not monitor. It flies stuff between here and there, that is all.”

He imagines that specific cargo routes will develop in Africa at around Eiffel Tower height in what he calls “the lower sky.” Unlike Google, he does not imagine that they will fly all around; it will not be Uber for stuff one can buy at CVS. “The routes will be geofenced: donkeys will only be able to fly in an air corridor about 200 metres wide and 150 metres high,” Ledgard wrote. “Busier routes will resemble a high-speed ski gondola, without cables or supporting structures.”

At the stops on the route, “every small town will have its own clean energy donkey station” that will “mix 3D printing and other advanced technology with low tech, presaging a Tatooine future where neural circuitry and simple materials will be matter-of-factly combined.”

Ledgard believes “there isn’t going to be enough cash for Africa to build out its roads.” Yet, in previous generations, good roads were an enabling condition for industrialization and realizing jumps in the standard-of-living. How might African nations and citizens experience greater prosperity? The only way, Ledgard has concluded, is through the air.

A decade traveling the continent for The Economist, reporting on everything from jihadis to the spread of cheap Nokia cell phones has convinced him that a technological paradox will permeate poor countries in the 21st century.

“A community will have access to a flying robot even though it will not have access to clean water, or security, or be able to keep its girls in school.”

This may sound absurd, but that doesn’t mean it won’t be the future we live.

* * *

Google has a specific vision for the future of self-flying vehicles, but its mere public entry into the field will catalyze all the efforts enumerated here from Matternet’s similar project to Ledgard’s radically different donkey vision. Google simply showing interest in flying drones legitimizes all these efforts by people who are trying to marshal much greater resources than they currently have to make their initiatives work.

Beyond the reputation boost, the unmanned cargo plane booster, Heerkens, hopes that Google will develop its program in a way that allows other companies to tap into its infrastructure. “The significance of what Google does, to me, is less in the vehicles they use here and now,” Heerkens said, “but the possibility in being a big organization of implementing the support infrastructure that’s needed.” For example, the detect-and-avoid systems will need to be certified, he believes, and Google could help governments figure out how to do so.

Matternet’s Raptopoulos wondered, too, whether they might not launch a service, but provide the cloud infrastructure for others to operate their own vehicles. “Google understands data infrastructure and mapping at the different levels better than almost anybody else. They may be thinking about an infrastructure play more than a service play,” said Raptopoulos, who had spoken with Teller about the project. “But this is all speculation.”

One area where Google will almost certainly have a major impact is in shaping the regulations that ultimately govern unmanned aircraft. “To a far greater degree than Amazon, Google has a history of working with policymakers and stakeholders on technology reform,” the University of Washington’s Ryan Calo, an expert on drone regulation, said. “Think net neutrality, fair use, privacy, and recently transportation. Adding Google’s voice could have a significant effect on regulatory policy toward drones.”

In Google’s case, that may mean they do what they’ve done with self-driving vehicles, where they hired Ron Medford, a former official at the National Highway Traffic Safety Administration, to lobby regulators on their behalf. Medford, backed by Google, have had a clear influence on legislative processes in California, Nevada, and other states where self-driving car laws have passed. In this case, Google could hire someone from the Federal Aviation Administration and perhaps make similar in-roads.

Teller confirmed that Google wants a seat at the regulatory table. “It’s gonna take conversations with the public and with regulators. But so far in the conversations we’ve had over the last two years, and more intensely over the last couple months with regulators, I’m cautiously optimistic that everyone wants the same thing,” he said. “Everyone wants the world to be a great place that’s safe and has the benefits of the technology with as little or no downsides as possible.”

Never were more Google-y words spoken.

Alexis Madrigal is a contributing writer at The Atlantic and the host of KQED’s Forum.