Inside X’s Mission to Make Robots Boring

Hi, everyone! A “decentralized autonomous organization” (translation: a group of crypto lovers) has decided to buy an original copy of the Constitution. Maybe they’ll add Satoshi’s signature? 

Image may contain Logo Symbol Trademark Text and Label
The Plain View

Every afternoon at 1:30, the robots file into the café. Gliding their way into the dining area on four wheels, the cadre of one-armed creatures weave through a maze of tables. Like hunters stalking prey, they swivel their binocular-sized heads, staring at their surroundings through an array of cameras and sensors. Additional vision is supplied by a Lidar sensor, similar to those on autonomous cars. If these robots had weapons, their appearance would incite panic. But instead, attached to the end of each of those robot arms is a custom squeegee.

X Robot sensors and camera

Five eyes: the array of cameras and sensors in the head of an Everyday Robot.

Photograph: Michelle Groskopf

These creatures are targeting tabletops. One of them will wheel up to a table and ponder for a few seconds to determine if people are seated; if so, it moves on until finding one that’s empty. After lingering for a second—maybe taking the algorithmic equivalent of a deep breath before the “Let’s do it” moment—the robot twirls and unfurls its limb, stretching the arm over the table to methodically cover the surface with a clear disinfectant. Then it withdraws the arm to squeeze out the excess fluid into a bucket on its base. Task completed, it moves on, seeking another table to swipe.

People finishing their lunch don’t even bother to look up. The robots have been doing this for weeks.

Five X Robots. Three of the robots have mop attachments.

Everyday Robots has built more than 100 robots in X’s Mountain View headquarters.  

Photograph: Michelle Groskopf

No, this isn’t a desperate attempt to address the labor shortage. It’s research by Everyday Robots, a project of X, Alphabet’s self-styled “moonshot factory.” The café testing ground is one of dozens on the Google campus in Mountain View, California, where a small percentage of the company’s massive workforce has now returned to work. The project hopes to make robots useful, operating in the wild instead of controlled environments like factories. After years of development, Everyday Robots is finally sending its robots into the world—or at least out of the X headquarters building—to do actual work. It’s enough of a milestone that they invited me to observe, two years after WIRED’s Tom Simonite last looked at the project. At that point, they had robots sorting trash into the proper recycling bin. Janitorial services represent the next, if not the final, frontier.

Attack of the squeegee robots: X’s fleet wipes tables in a Google cafe.Video: WIRED Staff
Portrait of Darcy Grinolds

Darcy Grinolds leads Everyday’s hardware reliability and design validation team.

Photograph: Michelle Groskopf

I kid, but this is serious stuff. Everyday Robots is trying to do two really hard things, a challenge so hairy that some question whether the effort is worth it. The first is credibly performing the tasks of human helpers. Everyday Robots lives on the razor’s edge of Moravec’s paradox, which states that it’s relatively easy for computers to perform difficult cognitive work and devilishly difficult to duplicate the functions of a two-year-old. Elsewhere under the Alphabet umbrella, robots navigate complicated traffic routes, drive automobiles more safely than humans, and become the champion of Go. In the Everyday Robots world, conquering a mundane task, such as crossing a cluttered room and opening a tricky door handle, is like winning the Super Bowl. The table wiping activity, for instance, isn’t just the swipe—it includes a whole suite of actions leading up to it. Take what happens when the path is blocked by a human or object. “The proper response for the robot is, OK, do I have enough space to gracefully move around that?” says Darcy Grinolds, who is the lead of the project’s hardware reliability and design validation team. “Or do I need to reroute myself around completely?”

The second hard thing the project is attempting to do is move toward that goal in such a way that it makes more sense, in terms of both economics and efficiency, to have a robot on hand than a bored and underpaid human.

Opening the door to a robot future.

Video: WIRED Staff

Google, and now X, has been obsessively pursuing this vision for more than a decade. Leading the Everyday Robots team is Norwegian-raised engineer Hans Peter Brondmo, an entrepreneur and engineer who joined X in 2015 and had to make sense of a cacophony of robotics acquisitions by the former leader Andy Rubin, who left the company under a cloud of sexual harassment claims. “Hans Peter was not the obvious choice,” says X’s CEO Astro Teller. “He cares about robotics, but he would be the first person to tell you he’s not a world-class roboticist. I picked him because he’s a world-class entrepreneur who really understands people. And he’s sort of a dyed-in-the-wool socialist—he comes from Norway!”

In an office he shares with a nonfunctional robot arm he built as a teenager, Brondmo explains that making an effective general-purpose robot became possible only with recent advances in machine learning. The engineers use machine learning to train the software to recognize objects and then run millions of simulations to compress weeks of testing into hours. This helps the lumbering robots in his lab to truly understand their environment, and build on that knowledge to accumulate a toolset that helps solve the inevitable dilemmas of coping in the wild. While Everyday Robots might not be as flashy as the dystopian androids in Boston Dynamics videos, they are optimized to get stuff done. (Alphabet once owned Boston Dynamics, but sold it off in 2017.)

Hans Peter Brndmo with two robot arms  one from a new robot and one from a robot.

Hans Peter Brondmo, general manager of X’s Everyday Robots, with two friends—a state-of-the-art robot (left) and the robot arm he built as a teenager in Norway (right).

Photograph: Michelle Groskopf

“Yes, you see really cool demos on YouTube of robots that can do backflips,” says Brondmo. “But those robots don’t know anything about the environment. You might say our robots are slow, but what they’re actually doing is fully autonomous. And they’re working in the world we live in, and they’re actually learning to do things, simple tasks that are getting more and more advanced,” he adds. “We’re bringing robots that can live and work alongside us into the world we live in, as opposed to us moving into the world they live in.”

portrait of engineer

Robot learning lead Mrinal Kalakrishnan trains the collective intelligence of the robot fleet .

Photograph: Michelle Groskopf

Everyday’s robot learning lead, Mrinal Kalakrishnan, showed me how his team trained a robot to push open a door, a task necessary for some of the robots’ jobs on the Google campus—for instance, going into conference rooms and determining via sensors if there are toxic levels of carbon. (One imagines robotic Covid monitors at some point in the future.) The door-latch training took under 10 hours. But once a single robot learns something, the knowledge is sent back to the whole fleet’s collective cloud intelligence, and thereafter all the robots can open doors with those latches.

On one hand, it’s astonishing to see these robots monitoring conference rooms and wiping tables. But then you ask yourself—why is Alphabet spending millions of dollars to perform chores that wouldn’t challenge a three-year-old? The corollary to Moravec’s paradox is Samuel Johnson’s unfortunate comparison of a woman preaching to a dog that walks on his hind legs: “It is not done well; but you are surprised to find it done at all,” said the 18th century writer and lexicologist. Sometimes at Everyday Robotics it is done well—maybe the most mind-boggling demo I saw on my visit was a choreographed robot dance where three robot units performed an eerily emotional ballet with the team’s artist in residence, Catie Cuan, a dancer pursuing a Stanford robotics PhD. Observing Cuan’s movements and coordinating their own actions via the cloud, the robots executed an intricate human-replicant interplay, while expending spare cycles to provide an improvised soundtrack made of a library of samples from the London Symphony Orchestra.

While executing Catie Cuan’s choreography, the robots respond to her movements and spontaneously generate a soundtrack from samples of the London Symphony Orchestra.

Video: WIRED Staff

X robot wiping a table where Benjie Holson sits.

Software engineer Benjie Holson, who grew up in a puppeteering family, appreciates a robot-swiped table.

Photograph: Michelle Groskopf

Two real tests need to be aced before everyday robots will be truly useful: They have to be cheap enough to be a cost-effective alternative to human labor, and they have to be flexible enough to cope with the near-infinite number of unplanned obstacles they will encounter in the chaotic reality that humans easily navigate. X engineer Benjie Holson—he grew up in a family of puppeteers, acquiring a perspective that he says is helpful in his current job—admits that at this point, if the robots cleaning tables were supplied with soy sauce instead of disinfectant, they would blithely soil the surfaces with it.

Getting to the point where robots can figure out if their squeegees are squirting soy sauce—and solving zillions of other unexpected real-world setbacks—is a Sisyphean journey. But it wasn’t long ago that robotic translation, or creative writing, seemed equally elusive. Assessing his team’s advances, Brondmo believes we’re on the cusp of breaking through. “That doesn’t mean that it’s not years away from being real,” he says, “but robotics are going to be a thing where the next generation is going to treat them like we treat these things”—he gestures to his phone. “Second nature.”

It may take a long time before all of us are dancing, so to speak, with robots on an everyday basis. But Brodmo and Teller are convinced that robots wiping off tables—and even bussing them!—will one day be common sights. When pressed for a business case for Everyday Robots, which I doubt will ever be cheap household items, Teller speculates that the average customer won’t buy them like Roombas, but instead arrange with a service provider to send out the bots to perform tasks, whether to clean corporate offices or help elderly clients access the loo. “Am I hoping that robots can help me to live more independently for longer?” asks Teller. “One hundred percent! I would be shocked if, in my dotage, I don’t end up that way.”

And the ghost of Samuel Johnson would be shocked if he did.

Image may contain Label Text Symbol and Sign
Time Travel

The quest to create a mobile robot that performs useful tasks goes back well over half a century. In my 1992 book Artificial Life, I describe how now-famed roboticist Rodney Brooks shared an office with Hans Moravec, whose name is immortalized in the paradox invoked above by the Everyday Robots team. Observing Moravec’s experiments helped inspire Brooks to create breakthrough technology like the Roomba vacuum cleaner.

Moravec was building a mobile robot. His goal was to place the robot on one side of a room and have it maneuver its way across the room, avoiding trash cans and desks. He approached the problem with standard assumptions. In order to negotiate its journey successfully, the machine would need a silicon “brain,” which would be the center of its cognition. In the memory of this brain, the robot would retain some sort of representation of the room. It would use vision sensors to “see” this room and compare it constantly with its mental representation of the area. And it would “know” things. It would know what its goal was. It would know what an obstacle was when it saw it. It would know how to move around obstacles. If such a wondrous consummation were ever to occur, it would know when it completed its task.

Moravec was extremely dedicated and clever, if a little odd. He lived in a makeshift warren in Stanford’s AI building, between the ceiling tiles and the roof. His robot was among the best and smartest autonomous robots the world had ever seen. Yet Brooks could not help wondering whether there might be a better way to go about making such creatures. “It would sit and compute for fifteen minutes and move a meter, then sit and compute for 15 minutes more,” Brooks recalls, with unseemly amusement. “That seemed long for me. I didn’t want a real slow robot. I wanted one that was faster to begin with. I wanted a robot to be in the world, with real people around.”

Image may contain Symbol
Ask Me One Thing

Steven (no, not me) asks, “Why can’t Congress legislate metaverse neutrality and interoperability? Just jump into the future rather than try to play catch-up with the present!”

Steven, thanks. Note that I am not calling you “Steve.” It’s annoying when people do that! Now to your question. It’s way too early to think about making laws to ensure neutrality in artificial worlds. There will be many metaverses, and in some of those, exclusivity might be a feature, not a bug. For instance, if a metaverse provides the experience of ancient Greece, it would be weird for people to drop in with their Star Wars–based avatars.

But you are probably more concerned with the prospect of one or two companies dominating the field with their proprietary metaverses, much in the same way that a handful of companies rule social media. That’s a real concern, but since we’re not yet sure that this whole metaverse thing will even happen, making rules now would be overkill. New statutes might even wind up favoring the currently entrenched social media giants, which might be able to lobby the prospective laws to their advantage. Ultimately, if the nightmare you fear comes to pass—a private company ruling a dominant metaverse—we have to hope that our current government regulatory apparatus, in the US and elsewhere, will kick in. The antitrust battle might even be fought in a virtual courtroom in ancient Greece!

You can submit questions to mail@WIRED.com. Write ASK LEVY in the subject line.

Inside X’s Mission to Make Robots Boring 1
End Times Chronicle

After a storm in Aswan, Egypt, swarms of deadly scorpions invaded homes and stung hundreds of residents. Ouch!

Image may contain Label Text Symbol and Sign
Last but Not Least

Finally, a break in Apple’s long-standing resistance to letting you fix your own iPhone.

A seriously unsung Black woman invented modern home security systems in 1966. She’d probably be horrified by what’s happened since.

A small cohort of “deplatformed” Substack writers are riding their cancellations all the way to the bank.

Meet Timothy Morton. English professor. “Poet-philosopher for our age of ecological crisis.” Explainer of “hyperobjects.” Reminder of why I quit graduate school.

Plaintext is celebrating Thanksgiving next week—enjoy the holiday and see you in December!

Image may contain Logo Symbol Trademark Text and Label

Don’t miss future subscriber-only editions of this column. Subscribe to WIRED (50% off for Plaintext readers) today.

If you buy something using links in our stories, we may earn a commission. This helps support our journalism. Learn more.


More Great WIRED Stories