These creatures target tablet computers. Someone will go up to a table and meditate for a few seconds to determine if people are sitting; If so, it will continue until an empty one is found. After waiting a second—perhaps taking an algorithmically equivalent deep breath before a “Let’s do it” moment—the robot spins and straightens its tip, extending its arm over the table to methodically cover the surface with clear disinfectant. Then he pulls the lever to squeeze out the excess fluid into a bucket on its base. Mission completed, you move on, looking for another table to pass.
People who finish their lunch don’t bother looking. Bots have been doing this for weeks.
No, this is not a desperate attempt to address the labor shortage. It’s research by Everyday Robots, a project of X Corporation, Alphabet’s “lunar shot factory.” The cafe’s test yard is one of dozens on Google’s campus in Mountain View, California, where a small percentage of the company’s huge workforce is now back in business. The project hopes to make robots useful, operating in the wild rather than in controlled environments like factories. After years of development, Everyday Robots is finally sending its robots out into the world – or at least outside of Headquarters X – to do real work. It was such a milestone that they invited me to keep an eye on it, two years after WIRED’s Tom Simonnett last looked at the project. At that point, they had robots sort the trash into the appropriate recycling bin. Janitorial services represent the next, if not final, frontier.
I’m kidding, but this is serious stuff. Everyday Robots tries to do two really hard things, a challenge so hairy that some wonder if the effort is worth it. The first is to credibly perform the tasks of human assistants. Everyday robots live on the edge of the shaver for Moravec’s paradox, which states that it is relatively easy for computers to perform challenging cognitive work and extremely difficult to replicate the functions of a two-year-old. Elsewhere under the Alphabet umbrella, robots navigate complex traffic routes, drive cars more safely than humans, and become a Go champion. In the world of everyday robots, overcoming a mundane task, like crossing a crowded room and opening a difficult door handle, is like winning the Super Bowl. The table scanning activity, for example, isn’t just scrolling – it involves a whole bunch of actions that lead to it. Take what happens when the path is blocked by a human or a body. “The appropriate response of the robot is, OK, do I have enough room to gracefully move around that?” says Darcy Greenolds, who leads the project’s hardware reliability and design validation team. “Or do I need to completely reorient myself?”
The second difficult thing the project is trying to do is move toward that goal in a way that makes more sense, in terms of economy and efficiency, to have a robot on hand than a bored, underpaid human.
Google, and now X, have anxiously pursued this vision for more than a decade. The Everyday Robots team is led by Norwegian-born engineer Hans-Peter Brondmo, a businessman and engineer who joined X in 2015 and had to come to terms with the idea of buying robots by former leader Andy Rubin, who left the company under a cloud of sex. harassment claims. “Hans Peter was not the obvious choice,” says Astro Teller, CEO of X. “He cares about robotics, but he’ll be the first person to tell you that he’s not a world-class roboticist. I chose him because he’s a global entrepreneur who really understands people. And he’s kind of a dyed socialist — he comes from Norway!”
In an office he shares with a nonfunctional robot arm he built as a teenager, Brondmo shows that making an effective general-purpose robot is only possible with recent advances in machine learning. Engineers use machine learning to train software to recognize objects and then run millions of simulations to compress weeks of testing into hours. This helps the holey robots in his lab really understand their environment, and build on that knowledge to assemble a toolkit that helps solve the inevitable dilemmas of adaptation in the wild. While the everyday robots may not be as flashy as the forlorn androids in the Boston Dynamics videos, they are optimized to get things done. (Alphabet once owned Boston Dynamics, but sold it in 2017).