Intelligent Machines

Google Hasn’t Given Up on Robots

Google is selling off a company working on the most advanced problems in robotics, but that doesn’t mean it doesn’t have robot dreams.

Mar 21, 2016

In 2014, Google went on a robot spending spree, buying a handful of companies working on various technologies to help robots see, walk, and grasp objects. It seemed that the company was intent on building advanced new robots that might transform factories and even our homes.

But last week Bloomberg reported that Google wants to sell the most striking of the companies it acquired, Boston Dynamics. The company’s impressive two- and four-legged robots were apparently too far from being marketable. Google has not given up on robots, but appears to have decided to be more realistic about what it can achieve.

“When you’re making a product, you have to pick what it’s going to do and a price point, and the target customer,” says Helen Greiner, who cofounded iRobot, a successful manufacturer of household robots such as the Roomba, and is now CEO of drone builder CyPhyWorks.

The latest four-legged robot from Boston Dynamics, called Spot.

Boston Dynamics is instead focused on inventing completely new capabilities for robots, primarily an uncanny sense of balance enabled by something called dynamic stability, which involves constantly moving to avoid toppling over. Figuring out how a technology that has just been invented can be used in a product can take years. “It’ll take a long time because they’re doing the research,” says Greiner.

Google is still developing technologies from the other robot companies it acquired, for example in a project using machine learning to teach industrial robotic arms to grasp objects more effectively. The companies acquired by Google in 2014 included Industrial Perception, which specialized in machine vision, and Redwood Robotics, which made advanced robot arms. However, similar work is being done by numerous research labs and robotics companies (see “This Factory Robot Learns Jobs Overnight”).

Greiner says work like this suggests that Google still wants to commercialize robot technologies. “I don’t see Google backing away from robotics at all,” she says. “They just might not have found a market for dynamically stable robots.”

Boston Dynamics, famous for its remarkably nimble and lifelike legged robots, was the standout among the companies acquired by Google in 2014 (see “The Robots Running This Way”). It isn’t hard to imagine that Andy Rubin, the executive charged with creating a forward-looking robotics division within Google, was as wowed as anyone else by the YouTube videos showing the Boston Dynamics two- and four-legged creations demonstrating incredible dynamic balancing tricks.

But Boston Dynamics is trying to solve problems that are considerably more challenging than any other of the other companies Google acquired, by designing robots to walk over unknown terrain. This might eventually prove useful for robots designed to work in messy human spaces, such as homes or hospitals, but it is currently very difficult to pull off.

Boston Dynamics was founded in 1992 by Marc Raibert, a pioneer in the field of legged robotic locomotion and previously a professor at both Carnegie Mellon University and MIT. Raibert had showed that robots could walk across uneven ground, or even treacherous terrain such as rocks using a neat trick: they keep moving in order to keep their balance. Raibert’s first robot was, in fact, a one-legged bot that kept itself upright by continually bouncing.

Despite being able to walk over mud, snow, or ice, Boston Dynamics’s machines cannot combine nimble locomotion with precise control. This much was clear when the company’s humanoid robot, Atlas, was involved with a DARPA challenge involving various straightforward everyday tasks (see “Why Humans and Robots Struggled with DARPA’s Challenge”).

That isn’t to say that Boston Dynamics isn’t making progress. The latest video from Boston Dynamics showed its newest humanoid robot moving through a warehouse and picking up boxes. On the other hand, the company’s robots cost millions of dollars each and require enormous amounts of power to perform their dynamic balancing. Continual refinement and economies of scale will be required to bring the cost and power demands down.

After Rubin left Google at the end of 2014, the robotics project apparently struggled with a lack of direction. Some suggest that Boston Dynamics, which is still located in Boston rather than on the West Coast, proved a bad cultural fit.

However, Raj Reddy, a professor of computer science and robotics at CMU, believes that the decision to sell the company probably has far more to do with commercial realities. “They bought all these companies because they thought this is going to be another major thing,” says Reddy. “I’m not sure you’re going to sell 10 million Big Dogs,” he says, referring to Boston Dynamics’s largest four-legged robot.

Reddy believes the technology developed at Boston Dynamics will live on at a company that has a longer-term vision. “The market for robotics companies is still hot,” he says. “People like Toyota, Honda, and Amazon are supposedly interested, and may be bidding a good price.”