These Robots Run, Dance and Flip. But Are They a Business?

By Cade Metz and Max Aguilera-Hellweg

Atlas 2.0, a mobile two-legged humanoid, at the Boston Dynamics lab and headquarters in Waltham, Mass. in September.CreditCreditPhotographs by Max Aguilera-Hellweg for The New York Times

Atlas 2.0, a mobile two-legged humanoid, at the Boston Dynamics lab and headquarters in Waltham, Mass. in September.CreditCreditPhotographs by Max Aguilera-Hellweg for The New York Times

WALTHAM, Mass. — Moving like a large dog, knees bent and hips swaying, the robot walked across a parking lot and into a rain puddle.

Boston Dynamics’s robot only looks like it thinks for itself.Published On

There, it danced a jig, splashing water across the asphalt. Then it turned and trotted toward a brick building, climbing over a curb and stopping within inches of a floor-length window. Pausing for several seconds, it seemed to eye its own reflection in the glass.

The scene was mesmerizing — so mesmerizing, it was easy to forget that a woman was guiding the four-legged machine from across the parking lot, a joystick in her hands and a laptop computer strapped to her waist.

The SpotMini is guided back to the human controlling it.Published On

The robot was called SpotMini. It was designed by Boston Dynamics, a company widely known for building machines that move like animals and humans. Thanks to a steady stream of YouTube videos from the otherwise secretive robotics lab, its machines have become an internet phenomenon.

But YouTube fame has not translated to very much revenue. In the coming year, Boston Dynamics, which was founded in 1992, plans to start selling the SpotMini, its first commercial robot. The mechanical dog would be a turning point for an outfit that has bewildered people with both its wondrous technology and its seeming lack of interest in making things someone — anyone — would actually want to buy.

Even now, it is not entirely clear what someone would do with one of these robots. That makes it hard to get past a question people have been asking about Boston Dynamics for years: Is this a business or a research lab?

“We think the technology has reached a point where it can be deployed productively,” Marc Raibert, the company’s founder, said during a recent interview inside his robotics lab, about 10 miles from Boston. “But we don’t know what the big application is.”

As the rest of the tech industry has focused on robotic cars and other contraptions that can navigate roads and warehouse floors, Boston Dynamics, which is owned by the Japanese conglomerate SoftBank, has plugged away at machines that can walk through the woods, into a rock quarry, across your home.

“These robots can climb stairs,” said Sangbae Kim, a professor of mechanical engineering at the Massachusetts Institute of Technology who is working on similar machines. “They can jump on a table.”

But if driverless cars are still years away from everyday use, walking robots are even further.

Though these machines are shockingly lifelike, they have limits. They can handle some tasks on their own, like spotting a curb and climbing over it. But when moving across unfamiliar spaces, like the parking lot outside the Boston Dynamics lab, they still need a human guide. In person, they stumble and fall more often than they do on YouTube.

Walking through the Boston Dynamics lab, Mr. Raibert, 68, wore bluejeans and a Hawaiian shirt, as he does nearly every day. He wants to build robots that can do what humans and animals can do. That was his aim in the early 1980s, when he founded the Leg Lab at Carnegie Mellon University in Pittsburgh. And it was his aim when he moved the lab to M.I.T.

In 2013, Google acquired Boston Dynamics during a broad push toward driverless cars and other robotics. But just four years later, Google, known for its “moonshot” bets on long-term tech projects, sold Boston Dynamics to SoftBank.

Mr. Raibert declined to say how much Google or SoftBank had paid for the company. He said both owners had provided ample funding for its research, and their involvement shows the potential of his work.

No machine comes closer to his vision than Atlas, a 165-pound anthropomorphic robot that can run, jump and even do back flips. Mr. Raibert would not let us shoot video of Atlas or other robots while inside the lab. But he did give a brief demonstration of the machine.

Like the SpotMini, Atlas is controlled by a joystick, a laptop computer and a wireless radio. When Mr. Raibert signaled for the demo, an engineer touched the joystick and the 165-pound robot crashed to the floor. Atlas is so large and so lifelike, you feel bad for it.

Mr. Raibert and his colleagues put the machine back on its feet and tried again. When the engineers touched the joystick, Atlas crashed to the floor a second time.

Eventually, the engineers got Atlas up and running — and sprinting and jumping and crashing headlong into an orange pylon. The excitement was mixed with discomfort when an engineer shoved its chest with the end of a broom handle and it stayed on its robotic feet.

“We subconsciously treat robots like living things — and I don’t think that effect will go away as we get used to the technology,” said Kate Darling, a researcher at the M.I.T. Media Lab who specializes in the ethics of robotics. “In some ways, we are biologically hard-wired to respond in this way.”

With its YouTube videos, Boston Dynamics plays with this phenomenon. The videos rarely reveal the human operators who guide the machines.

SpotMini can open a door with an optional attached arm, then kick the door wide open with its hind legs as it passes through.

SpotMini is smaller and cheaper and has better balance than Atlas. It can carry (small) items on its back, and it can open doors (provided the doors have the proper handles). This requires an extra limb that attaches between its shoulders.

Mr. Raibert says that extra limb’s resemblance to a head and neck is unintended. He said he was surprised when YouTube watchers described one of the company’s robots as “biting” a cinder block. Without a head and neck, SpotMini is perhaps less menacing — but also less charming. In recent months, the machine has become a regular guest on the tech conference circuit, appearing with the likes of Amazon’s boss, Jeff Bezos.

The sales plans for SpotMini seem vague. It will be priced like a car — cars, Mr. Raibert added, have a wide range of prices — and it will be sold to businesses like construction companies. He talked in general terms about the machine’s lugging stuff across rough terrain or into places unsafe for humans.

As the technology improves, it could be a delivery robot, walking packages down the street, up your front steps and onto your porch.

Mr. Raibert showed how a SpotMini could serve as a kind of automated security guard. Boston Dynamics has created a three-dimensional map of its lab, and the SpotMini can use this map to navigate the lab on its own. It could, in theory, patrol after everyone else had gone home.

SpotMini inspecting an object under a desk during a demonstration at Boston Dynamics.

When Google acquired Boston Dynamics, it ended the company’s military contracts. Now that the company is owned by SoftBank, Mr. Raibert said, it could also return to military work.

Boston Dynamics does seem to have an owner willing to wait until the business gets figured out. And with a $100 billion investment fund, a partnership with the crown prince of Saudi Arabia, Mohammed bin Salman, SoftBank is betting on technologies that require years of work. Before buying Boston Dynamics, it acquired a French robotics company, Aldebaran, which also is working on machines that are a long way from completion.

Robots are a focus of Japanese tech researchers. SoftBank is already offering a machine, called Pepper, that automates parts of customer service at retail stores. Schaft, which is owned by Google despite reports of a sale to SoftBank, is exploring machines similar to those at Boston Dynamics. And the Toyota Research Institute is working toward robots that can serve Japan’s rapidly growing population of retirees.

At Boston Dynamics, Mr. Raibert wants to sell robots to businesses, governments and all sorts of other customers.

He calls himself “a lifer” in his quest to build machines that can do everything animals and humans can do. And if that means finding a way to make money from his experiments along the way, so be it.

Cade Metz is a technology correspondent, covering artificial intelligence, driverless cars, robotics, virtual reality, and other emerging areas. He previously wrote for Wired magazine.  @cademetz

A version of this article appears in print on , on Page B1 of the New York edition with the headline: For Sale: One Robot In Search Of a Job. Order Reprints | Today’s Paper | Subscribe