The Autonomous Selfie Drone Is Here. Is Society Ready for It?

It’s 2035, the Second American Civil War has been won by the other side, and you find yourself in a heap of trouble with Attorney General Logan Paul. (The future is very troubling.) He has dispatched an all-seeing eye-in-the-sky to tail you, an agile flying machine equipped with 13 cameras and a top speed of 25 mph.

The drone knows your face, your gait and your clothing. It hovers persistently behind your back, moving when you move, stopping when you stop, resisting every effort to shake it. You run into the woods, but you still can’t lose it.

So now what? Clip this article and save it as a guide for surviving our airborne future. In a woodsy park in San Francisco last week, I had an encounter with just such a self-flying drone, and I found only one trick for escape. Hint: It involved the indignity of repeatedly running around a tree.

As the hapless chump in its cross hairs, I will tell you this: Being tailed by a 13-eyed flying machine has a way of focusing the mind.

The drone chasing me, the R1, was created by a startup called Skydio; it sells for $2,499 and will begin shipping to customers in two to three weeks, the company says. It is the closest thing to a fully autonomous drone you can buy today.

Autonomous drones have long been hyped, but until recently they’ve been little more than that. The technology in Skydio’s machine suggests a new turn. Drones that fly themselves — whether following people for outdoor self-photography, which is Skydio’s intended use, or for longer-range applications like delivery, monitoring and surveillance — are coming faster than you think.

They’re likely to get much cheaper, smaller and more capable. They’re going to be everywhere, probably sooner than we can all adjust to them.

Most consumer drones rely on some degree of automation in flight. DJI, the Chinese drone company that commands much of the market, makes several drones that can avoid obstacles and track subjects.

But these features tend to be less than perfect, working best in mostly open areas. Just about every drone on the market requires a pilot.

“Our view is that almost all of the use cases for drones would be better with autonomy,” said Adam Bry, Skydio’s chief executive.

Skydio was founded by Bry and Abe Bachrach — who met as graduate students at the Massachusetts Institute of Technology and later started Google’s drone program, Project Wing — along with Matt Donahoe, an interface designer.

In 2014, with funding from the venture firm Andreessen Horowitz, the company began working on what would become the R1. Skydio has since raised $70 million from Andreessen and several other investors, including Institutional Venture Partners, Playground Global and basketball player Kevin Durant.

Skydio’s basic goal was a drone that requires no pilot. When you launch the R1 using a smartphone app, you have your subject stand in front of the drone, then tap that person on the screen — now it’s locked on. You can also select one of several “cinematic modes,” which specify the direction from which the drone will try to record its subject. (It can even predict your path and stay ahead of you to shoot a selfie from the front.)

After takeoff, it’s hands off. The drone operates independently. In the eight-minute flight I saw — through a wooded trail sparsely populated with runners and dogs — the R1 followed its target with eerie determination, avoiding every obstacle as naturally as an experienced human pilot might, and never requiring help. It lost its subject — me — only once, but I had to really work to make that happen.

Watch a clip, recorded by the R1’s main camera, showing the drone following Bry. When he runs, it immediately follows him, ducking behind trees, speeding up to catch him, then slowing when he slows. Much of the rest of the clip goes like that; every step he takes, the R1 is watching him.

Time for some caveats: Skydio’s technology is far from perfect. It doesn’t work well in inclement weather or at night. It also requires a very high-powered processor, which gobbles up battery life; the R1 gets 16 minutes per flight, compared with around 20 for competing drones (but it will ship with two batteries, allowing for another flight after a quick swap out).

Skydio’s drone is also entering a crowded marketplace that hasn’t been kind to new players. A parade of drone startups have gone belly up in the last couple of years, unable to compete with DJI’s technical innovation and manufacturing scale. Some of the most spectacular blowups, like that of the drone startup Lily, involved overpromising and never delivering on autonomous features. Skydio’s founders say they’re keen to avoid that mistake. The company showed me the assembly line in its headquarters in Redwood City, California, where teams of technicians are finishing production-ready R1s by hand. The small-scale operation is meant to ensure quality, but it comes at a price. DJI’s latest feature-packed drone, the well-reviewed Mavic Air, sells for $800, a third of the price of the R1.

“I know this technology is so much smarter, but I’m just not sure it’s enough to overtake the behemoth that is DJI,” said Sally French, a journalist who covers the drone industry at her site, The Drone Girl. She was also shown a preview of the R1 last week.

Yet whatever happens to Skydio as a company, its flavor of autonomous tech seems likely to become ubiquitous. How Skydio has achieved autonomy is, at bottom, a marvel of software rather than hardware, and the software is likely to get cheaper and better quite quickly.

There are two basic ways for computers to process the visual world. They can use cameras alone, or they can also use depth sensors, like lasers or radar, that precisely determine where objects are in space.

Most self-driving car systems use expensive laser sensors, known as lidar, a spinning disk that sits on the car’s roof like a propeller beanie. Lidar was the key technology at issue in Waymo’s now-settled trade secrets lawsuit against Uber. The trial surfaced a note from Travis Kalanick, Uber’s former chief executive, that underscored lidar’s importance: “Laser is the sauce,” he wrote.

But to Skydio, laser wasn’t the sauce. Lasers aren’t only expensive; they’re also heavy and bulky. (That’s why Elon Musk also isn’t a fan.) Cameras, meanwhile, are plentiful and cheap. Several years ago, Skydio took a gamble: It would use only cameras, arranged in eyelike pairs on every side of the drone, and hope that it could get laserlike accuracy using ever-improving artificial intelligence techniques.

“A bird doesn’t need lidar to fly,” Bry told me.

The scheme worked better than the team was expecting. Because an artificial intelligence technique known as deep learning has been improving much more quickly than insiders predicted, Skydio said, it is further along at key perceptual tasks than it had hoped to be, and Bry said the tech was still advancing. What this means is ubiquity. As I watched the R1 tail Bry, I played the scene forward in my mind: What happens when dozens or hundreds of runners and bikers and skiers and hikers and tourists begin setting out their own self-flying GoPros to record themselves? Our society has proved in thrall to photography; if you can throw up a camera and get a shot of you reaching the summit, who’s not going to do it?

Even regulations may not be ready for the R1. Colin Snow, a drone industry analyst, pointed out to me that federal regulations require users to keep drones in their line of sight — and a drone that follows you is technically not in your line of sight.

Bry disputed this characterization. He argued that regulations allow for momentary periods when the drone is not in the line of sight, as long as the vehicle does not interfere with manned aircraft.

As for other social disruptions that autonomous drones might cause, Bry was guarded. “Obviously, we tell our users to be responsible,” he said.

Which brings me to the escape plan. How do you lose an R1 that has been told to follow you? Here’s what worked for me: Find a tree. Run around it very precisely, just fast enough to stay exactly opposite the drone, so the tree trunk blocks its view of you for seconds at a time. Do it just right and the bird will lose you, stopping in its tracks, confused.

But I don’t imagine this trick will work for long.