Spot, the Internet’s Wildest 4-Legged Robot, Is Finally Here

In a cavernous, half-built structure outside of Boston, I stand at the edge of a pit watching YouTube’s most famous robot do my bidding. A flick of a joystick with my left thumb sends the quadruped forward and backward and side to side, while my right thumb turns it left and right, motors whirring with each deft step. The machine clambers over rocks and tackles crumbling inclines with ease, sometimes sinking its feet into the dirt, nevertheless carrying on. Carelessly I steer it toward a pipe sticking out of the ground, and the robot trips over it. I react with momentary alarm, but Spot catches itself and continues on its way, as if it was nothing.

For years Boston Dynamics has been posting YouTube videos teasing its scary-cute four-legged robot—making it open doors for its friends and fight off stick-wielding humans. Today, finally, it is announcing that it’s unleashing Spot (formerly known as SpotMini) into the market. Spot is so unprecedented and so befuddling in its deftness that even Boston Dynamics isn’t totally sure what the robot will be good for. What is clear, though, is that Spot’s introduction to the workforce marks a unique moment in labor, the start of an era in which humans work more intimately with advanced machines.

“We’re going to work together with our customers to figure out what robots are good for,” says Marc Raibert, who heads BD. “It’s not like they can do every possible thing, but that doesn’t mean they’re not really good at some things.”

A bit of disappointing news for you right away: BD is leasing rather than selling Spot, and the company is being choosy about who gets to deploy the machine. It’s talking with energy companies, for instance, who might want Spot to inspect infrastructure, as well as with Cirque du Soleil, which is exploring Spot’s potential as an entertainer.

But in my visit to that Boston construction site, BD was emphasizing a different use for Spot: as a sort of patrolbot that can continuously inspect a space. On a typical job site, a manager might be able to peruse and document the whole operation once a week to make sure all the components are coming into place properly, and that crews aren’t, say, running a duct where there’s supposed to be a supporting column. But if you loaded out Spot with a 360 camera, it could do the rounds every night, because once it builds an initial 3D map of the environment, during each subsequent visit it can detect changes. It can operate in the rain and withstand clouds of dust. Should it end up turtled on its back, it does some coordinated leg wiggling to flip itself back onto four feet. Because it conveniently lacks lungs, it can venture into hazards like asbestos-tainted buildings, where it’s expensive to send humans in hazmat suits.

Spot equipped with a lidar unit

Photograph: Matt Simon

How much exactly Spot will run you, BD isn’t saying. “It’s about the price of a car,” says Michael Perry, vice president of business development at Boston Dynamics. “How nice of a car depends on how many robots and how long the lease period is.”

LEARN MORE

The WIRED Guide to Robots

As with a car, the price will also depend on the options you select. Spot comes standard with five pairs of stereo cameras positioned around the machine. Its front cameras scan the ground ahead for obstacles and decide whether to climb over them or dodge them altogether—anything under 30 centimeters tall gets summited; anything higher it will automatically avoid, even if you joystick the robot right at a wall. (In addition to the joysticks, Spot’s controller has a screen you can tap on, which shows a feed from the robot’s forward-facing cameras, to send the machine to waypoints. It’s a bit like teleporting up and down streets in Google Street View.) But you can add additional sensors on its back, like a lidar laser scanner for mapping an environment in high detail.

Spot is in a sense modular, in that a given set of sensors will qualify it for a certain job. So to deploy the robot, a client has to work closely with BD. “We’re not putting it on Amazon and saying, ‘Buy this,’” Perry says. Part of the trick is that something like a lidar unit requires additional compute power attached to the back, and all that extra weight muddies the control algorithms: Just as a horse moves differently with a human on its back, Spot must adapt to a heavier payload to maintain its famous deftness. Its different configurations also affect its battery life. Customers using the base robot can expect about 90 minutes of walking time, and to keep the thing running an operator can easily swap out the battery in its belly.

Video: Matt Simon

Human workers, too, have to adapt to Spot. “The robot has a high degree of autonomy, but it’s not like you can just put the robot into a space and say, OK, now scan everything,” says Perry. You still need a human around to make sure the robot can navigate an environment. On a construction site, for example, you’d have to first use a controller to maneuver Spot around the work site so it can build a map of the space. “That really contradicts a lot of people’s expectations, particularly after seeing some of our YouTube videos.”

Video: Matt Simon

Meaning, your expectations for what Spot can do are probably wrong. What you don’t see in the videos of Spot opening doors and fighting off humans are the outtakes, the many times that the robot fails. These machines actually rely heavily on humans to be useful. So when leasing Spot, BD has to sit down with a customer and make it plain that this ain’t YouTube anymore: Spot has a particular set of skills and shortcomings that human workers need to accommodate.

That and Spot might not even be the right tool for your needs—if you want to remotely monitor five gauges, just train five stationary cameras on them. And maybe you do need a robot, just not something as advanced as Spot. Drones can map environments too, and a wheeled robot like Knightscope’s security robot will work fine for flat environments without stairs. “We don’t want to mislead people into getting a solution they don’t actually need,” says Perry.

This budding relationship between sophisticated robots and human workers is as much about getting the machines to adapt to us as it is about us adapting to them. Because the tendency, especially for something that moves in such an animal-like way as Spot, is to over-ascribe agency to robots, when they’re still very limited.

Video: Matt Simon

“People will often name or talk to the robots they work with, even though they know that they’re just machines,” says MIT roboticist Kate Darling, who studies human-robot interaction. “Spot has such a biologically inspired design that it’s hard to imagine people won’t treat it a little bit like a pet.” But a pet this is not—Spot is a tool, and human coworkers must treat it as such.

The neat bit is that as we develop this relationship between humans and machines, we can tailor machines to fit specific niches, just as natural selection has molded the morphology of species. And at BD, what they learn developing Spot can be put to use on another robotic system. BD’s Atlas robot, for instance, walks on two legs but can inform the control mechanisms of Handle, a robot that balances on two wheels. As BD continues to refine Spot, the company can start thinking about designing new morphologies for new tasks.

“I could easily imagine us designing a variant on Spot that is larger and stronger, like you might find outdoors, more like a bulldozer, if it turned out there was enough interest in that,” says Raibert. “It would be an engineering project, but a lot of the intelligence and functionality that is in Spot would translate directly into a form like that.”

This is the vision of Boston Dynamics’ budding robot empire: R&D the hell out of the things, like they’ve been doing for nearly three decades, and use those discoveries to refine a range of robots that fit particular niches, whether it’s Handle scurrying around warehouses delivering boxes, or perhaps a humanoid like Atlas one day lifting elderly folks in and out of bed. In this paradigm, no robot evolves in isolation. And as these things get smarter, they can begin to learn on their own to, say, manipulate certain objects, then share that knowledge with each other.

Read More