Not many robotics companies can boast legions of fans online, but not many robotics companies make robots quite like Boston Dynamics.
Each time the firm shares new footage of its machines, they cause a sensation. Whether it’s a pack of robot dogs towing a truck or a human-like bot leaping nimbly up a set of boxes, Boston Dynamics’ bots are uniquely thrilling.
They’re also something of a Rorschach test for our feelings about the future, with viewers either basking in the high-tech splendor or bemoaning the coming robo-apocalypse. And when a parody video circulated last month showing a CGI “Bosstown Dynamics” robot turning on its creators, many mistook it for the real thing — a testament to how far the company has pushed what seems technologically possible.
But for all its engineering prowess, Boston Dynamics now faces its biggest challenge yet: turning its stable of robots into an actual business. After decades of kicking machines in parking lots, the company is set to launch its first ever commercial bot later this year: the quadrupedal Spot. It’s a crucial test for a company that’s spent decades pursuing long-sighted R&D. And more importantly, the success — or failure — of Spot will tell us a lot about our own robot future.
Are we ready for machines to walk among us?
Talk to anyone in the robotics industry and they’ll sum up their sector with a three-word phrase, honed by years of trial and error: robots are hard.
The sector is notoriously unforgiving, with startups and established companies often collapsing with little warning. Just last year, three robotics companies folded in the space of a few months. In response, industry veteran James Kuffner of the Toyota Research Institute summarized the challenges of building robots on Facebook. “It requires significant funding, committed leadership, highly skilled staff, resources, and infrastructure, and an excellent product and market strategy,” wrote Kuffner. “Not to mention flawless execution.”
Boston Dynamics’ robots seem flawless, but that’s partly because they’ve never had to operate in the hurly-burly of commercial environments. Since its founding in 1992, the company has relied on deep-pocketed patrons like the Department of Defense and Alphabet. Its earlier life was shaped by government contracts, and in 2013 it was bought by Google’s parent company, as part of an abortive attempt for the search giant to enter the robotics industry.
Boston Dynamics CEO Marc Raibert tells The Verge that those years of contracting and research were necessary to bring the company to its current stage of development.
“We’ve been an R&D company for a long time, working on pushing the envelope [and] making robots that try to live up to people’s idea of what a robot should be,” says Raibert. “And it’s natural … that as we do that R&D it makes robots more and more useful, and it makes it obvious to us that, ‘Oh, this thing could be used and commercialized.’”
Pentagon contracts gave Boston Dynamics the time and space needed to develop cutting-edge legged robots like the pack-mule BigDog (eventually rejected by the military for being too noisy), but they’ve not yet led to a salable robot. Instead, the company has consistently impressed backers by giving machines a trait that’s eluded them for decades: mobility.
True mobility is something beyond the ken of most machines, explains Hod Lipson, a professor of engineering at Columbia University. “We think that playing chess is a big deal but just walking around, coordinating hundreds of muscles is an incredible accomplishment,” Lipson tells The Verge. “Robots, in the most part, are still very clumsy. The smallest physical obstacle bewilders them.”
Consequently, most machines used in factories and warehouses today are huge, static, and unintelligent things: designed to stay in one place and perform repetitive tasks. The robots of tomorrow, by comparison, will be agile and dynamic; capable of working alongside humans and reacting to changing environments and behavior. Incidentally, that’s why Boston Dynamics is so fond of pushing and shoving its robots in videos. There’s nothing like a swift kick to the ribs to prove that a robot can cope with physical uncertainty.
Boston Dynamics has been “trying to break this invisible boundary” of mobility for decades, says Lipson. But as a result, salable applications “almost seem like an afterthought.”
It’s an assessment with which many in the industry agree. “They were on the government dole then on the Google dole,” Erik Nieves, founder of automation company Plus One Robotics, tells The Verge. “They had no real mission: just be awesome! But they’re already awesome.”
The trigger for commercialization seems to have been the company’s acquisition in 2017 by Japanese tech giant SoftBank. Raibert says commercialization was always the end goal, but that access to SoftBank’s significant resources have allowed the company to kick its production of robots into a higher gear. “That’s one of the ingredients,” he says.
Softbank is an “unabashedly commercial company” and will want to “get a return on their investment,” Nieves says. Notably, the Japanese firm’s other bets in robotics — which include Aldebaran, makers of the Pepper robot, and Fetch Robotics, which does warehouse automation — have been selling robots in commercial settings for years.
As well as making Spot into a salable robot, the company has also bought logistics startup Kinema Systems to pave the way into warehouse automation. Boston Dynamics is already selling robots it acquired with this purchase, giving it a foothold in the new industry.
As part of this new focus, Raibert has become a familiar figure on the tech conference scene. Appearing onstage in his trademark Hawaiian shirts, he regularly wows audiences with tech demos, directing Spot to jump, trot, and dance, like a robot ringmaster.
Raibert’s big promise is that the Spot will become the “Android of robotics” — a customizable platform that other companies can build on to meet specific needs. “We specifically designed it as a platform so it can be customized for lots of different users,” says Raibert. “In very short order, we’ll have a number of different attachments that can be used to customize the robots.” (Though the actual launch date and price have still yet to be confirmed.)
So far, these payloads include robotic arms capable of grabbing and manipulating objects; sensor arrays including thermal and 360-degree cameras; as well as radio units, so Spot can become a mobile relay for communications. This flexibility means Boston Dynamics plans to sell and lease Spot for a wide array of tasks: everything from surveying construction sites and industrial buildings to package delivery and security applications.
Is there anyone Boston Dynamics wouldn’t sell Spot to? To law enforcement or the military, for example? Raibert doesn’t rule it out. “We’re enthusiastic about responsible use of the robot,” he says. “I think you’re asking a tough question because there’s so many edges on it.”
There’s certainly a market for these applications. Companies like Knightscope already offer robot security guards that patrol spaces like parking lots and malls. And while these bots are cheaper than humans (Knightscope’s cost roughly $7 an hour), they’re limited by their wheeled designs. Curbs are a problem and stairs an impossibility. A Knightscope robot demonstrated these flaws memorably in 2017 when it nose-dived into a fountain. It’s the sort of image that looks perfect in a pitch deck, with Spot on the next slide, trotting happily over obstacles.
At this point, Boston Dynamics’ robots look less like a single product and more like a vision of the future. Legged robots have been difficult to build for decades, but advances in a range of connected fields — including sensors, motors, control software, and machine vision — are making them viable for the first time.
Raibert says the difference between Spot and earlier legged robots is night and day. “It’s a tribute to what we’ve learned over the years in sensing the terrain, in balancing the robot, in controlling it,” he says. To reach this level the company has leaned on what Raibert calls “low-level AI.” That means artificial intelligence control systems that are responsible for keeping the machines upright and balanced in all situations. Actually telling the robot where to go and what to do is left to humans, who control Spot using a modified gaming tablet.
Legged designs have many natural advantages, says Lipson. They’re easy to balance and hard to knock over. They work in a wide variety of environments and are supremely adaptable. “This is why nature has so many legged machines,” says Lipson. “It’s a very versatile platform … I believe this will be the primary platform for future robotics.”
Boston Dynamics isn’t the only firm that’s sees this potential. In the years the company has spent honing its designs, numerous competitors have sprung up with similar products. These include Laikago, a quadrupedal robot designed by Chinese firm Unitree Robotics; the Vision and Wraith series made by the Philadelphia-based Ghost Robotics; and ANYmal, created by ANYbotics, a company spun out of ETH Zurich University in Switzerland.
ANYmal is particularly similar to Spot, able to work indoors and outdoors with a range of add-ons. Co-founder Péter Fankhauser tells The Verge that the company has already started selling its robots (though he won’t say for how much), and recently demoed one of its machines surveying an offshore energy platform in the North Sea.
It’s the perfect showcase for technology, says Fankhauser. Not only can a legged machine navigate the tight corridors and stairs of this sort of industrial environment, but sending a robot to remote locations means one less human stuck in the middle of nowhere. “These are dangerous jobs, remote jobs, and hence expensive jobs,” he says. “The business case is very, very clear.”
In Fankhauser’s view, it’s good news that more companies are launching legged robots. It creates a more varied market, he says, and gives customers greater confidence. “Companies would be afraid to buy these things if there was only one supplier.”
Jeff Burnstein, president of the Association for Advancing Automation, agrees that it’s a “good sign” that more companies than just Boston Dynamics are involved in this sector. But he tells The Verge that it’s hard to predict whether the bots will really take off given that the hardware is so new. “We just haven’t had many of these products on the market before.”
While the case for Spot seems largely optimistic, it might prove trickier for Boston Dynamics to get its robots running in a more banal, but potentially more lucrative, setting: the warehouse.
After the company bought Kinema Systems, it started selling the startup’s Pick robot, a static industrial arm equipped with pneumatic suckers that uses deep learning to see the world around it. (The arm itself is made by Japanese robotics giant Yaskawa.) This year, it showed how its wheeled robot Handle, which uses a counterweight to balance a single picking arm, might be incorporated into the same setting; depalletizing goods and stacking boxes.
Selling Pick seems straightforward: Kinema Systems proved the tech’s utility before the company was acquired. But incorporating Handle into this same environment is much more ambitious — more so than Spot. Boston Dynamics is presenting the robot as something closer to a direct stand-in for humans: a machine that’s able to navigate a warehouse as easily as a person. That means it could potentially be slotted right into a company’s workflow, rather than asking a firm to reorganize their factories or warehouses.
For Nieves, whose company also automates logistics and warehouses, it’s a tough job. He says Handle’s design is pure Boston Dynamics: graceful and agile, “a beautiful piece of engineering.” But it will be more expensive than other options, both human and mechanical, he warns, and it’s hamstrung by its design in a number of ways.
For a start, it’s untethered, meaning it would rely on battery power. That would necessitate companies buying multiple units in each location, so some could charge while others worked. Its pneumatic gripper is also only able to grab certain goods, making it tricky to use for a lot of common warehouse tasks, like loading or unloading a truck.
“There is no way robots can do that today,” says Nieves. “You are playing Tetris in 3D in real time and half of the stuff you’re having to load isn’t even rigid.” He says: “I’m convinced that Handle as you see it today will not see the inside of a true warehouse … They’ve got work to do yet in how they bring this to market.”
Raibert says the technology is still being developed, and that the company has no timeline for when Handle might go on sale. But he says the opportunity for warehouse automation is “massive.”
“You go look at logistic activities around the world and it’s essentially unautomated. People are distracted by thinking that Amazon has their warehouse entirely automated, but they just have one or two tasks automated,” he says. Having truly mobile, dynamic robots like Handle would change that. “There’s a lot of low-hanging fruit there,” says Raibert.
Like the commenters underneath Boston Dynamic’s videos, it’s hard not to see the company as a litmus test for what the future of automation will look like. And that’s not at all unfair — the future of automation really is uncertain.
Robots are becoming more common in everyday settings, but experts worry they’re not doing a very good job. In a paper this year, economists Daron Acemoglu and Pascual Restrepo — two of the most respected researchers in the field — warned of a phenomenon they dubbed “so-so” automation, when the technology used to replace people doesn’t offer any actual benefit to the economy. Automated factories don’t get faster or more productive; they just swap human labor for their machine equivalent, with the benefits of this exchange accrued by those in charge.
We’re already seeing this dynamic in play in parts of the workforce. In Walmart, for example, where robots are taking on mundane tasks like scanning shelves, employees say the machines don’t make their jobs easier. In fact, they make them harder, because of the extra work needed to manage the bots. In Amazon warehouses, robots are taking on more jobs, but as Raibert says, it’s still only partial and in the interim, humans are treated more like machines.
Acemoglu and Restrepo warn that if this trend continues and new job opportunities are not created, life will only get harder for the working people — something we’re already seeing with the rise of precarious jobs and wage stagnation.
It’s this depressing, deflating context that makes Boston Dynamics’ robot so exciting. Rather than making the same old trundling bots that fall over and wait for a human to pick them up, the company seems to be leap-frogging “so-so” automation into something more technologically advanced. At least, that’s what the company’s videos show. Now it’s Boston Dynamics’ time to prove that its robots are ready to leave the lab and head out into the world.
“I think robots are going to affect peoples’ lives in a good way. I think it’s going to increase productivity, I think it’s going to release people from dull, dirty, and dangerous [jobs],” says Raibert. “I would hate to see the great opportunities in a technology like this missed because of fear of what the downsides may be.”