Rod Brooks and Rethink Reveal an Industrial Robot for the Masses
“Want to see some robots?” says Rodney Brooks.
Yes I would, I say. And so would the rest of the world.
Brooks is a renowned roboticist and artificial intelligence expert, the founder, chairman, and chief technology officer of Boston-based Rethink Robotics. In a previous life, he was an MIT faculty member for 25 years, and he co-founded Bedford, MA-based iRobot (NASDAQ: IRBT), the maker of Roomba vacuum cleaners and PackBot military robots. He left his post as CTO of iRobot in 2008 to pursue his current venture.
With Rethink, Brooks (an Xconomist) has been sitting on a secret for a long time—namely, what the company’s robots look like, and exactly what they can do. But today is Rethink’s coming-out party, in preparation for shipping product next month. This is where top, cutting-edge robotics expertise meets big business. And this is what $62 million in venture funding buys you. (That money comes from Amazon.com’s Jeff Bezos, Charles River Ventures, Highland Capital Partners, Sigma Partners, and Draper Fisher Jurvetson.) Clearly, the company is one of the biggest technology bets in Boston—or anywhere, for that matter.
But first, why the recent name change to Rethink? I have been on record saying I preferred the old name, Heartland Robotics. Well, it turns out that “Heartland” technically means the interior of the country, so states on the coasts would be excluded, says Brooks. Heartland also doesn’t translate well overseas, he says (though I think it sounds pretty good in his Aussie-Boston accent). “Rethink” seems to encompass the bigger picture of what the startup is trying to do: revitalize the manufacturing industry with inexpensive and intelligent robot helpers.
Brooks leads me to an open area of the company’s headquarters where a half-dozen robots are working on benches, some of them connected to computers and various debugging tools. Engineers are tinkering furiously around us. “This is Baxter,” Brooks says, stopping in front of one of the robots. (The name is an archaic form of “baker” and strikes the right occupational tone, he says; it was one of hundreds of names suggested by Rethink’s employees.)
Baxter has an LCD screen for a head, a quad-core PC box in its torso, and two large, red, human-like arms. Brooks calls them “Olympic swimmer” sized, and they have shoulder, elbow, and wrist joints. If you look closely, they are not left and right arms—they’re both right arms (simpler to program and outfit them, I guess). The robot has five cameras—one in its head, two in its chest, and one in each wrist looking down at its workspace. It also has a sonar system up top to detect people and objects around it. In total, the robot weighs about 165 pounds and can be bolted to any work platform; it also has a base with wheels, so it can be moved around a factory floor, but doesn’t get around by itself.
You might think of this as the world’s first mass-produced, commercial humanoid robot—though it doesn’t actually look very human. And that’s not the point anyway. The point is to do manual tasks that will help factory workers in assembly-line-type jobs be more efficient. To do that, the robot needs dexterous hands and a simple user interface. And it needs to be cheap (for an industrial robot): Baxter will sell for $22,000, not including its base and hands, but including a software subscription and warranty.
Brooks shows me a demo (see video, below) of how it works. Using a two-fingered, pincer-like hand—the hand can be swapped out for others of different shapes and configurations, depending on the task—the robot learns to pick up a widget on the bench and put it down somewhere else. The task can be done with or without computer vision; using the cameras takes a little more processing time. Brooks starts by selecting a few operations using a combination of buttons and rotary selectors on the robot’s arm to navigate a menu on the display. He physically guides the robot’s arm above the widget and presses a button to close its fingers around the part to pick it up; then he guides it to a spot a few feet away where he opens Baxter’s fingers to put the part down. After a little training, the robot is able to repeat the sequence on its own. (The video cuts off just as Brooks is demonstrating the object recognition capabilities of Baxter’s computer vision system.)
Here’s what’s interesting. If you change the task a little—move the widget around on the table, say—the robot will grope around until it finds the part (or doesn’t find it, as it were). If you get in the way of the arm—Brooks put his head in its path to demonstrate—the arm stops abruptly. Its control system has a force field that prevents the robot from bumping into objects around it, including itself. And its joints are soft and springy: they are powered by what are called series elastic actuators, a technology licensed from MIT, so they have some give to them, instead of being rigid (and potentially dangerous) like traditional industrial robots.
And that’s the big idea here: Baxter is a new breed of industrial robot that’s meant to be accessible to the masses. A factory worker could set up the robot to pick parts out of a box, or place them into a box, or take them off an assembly line. The robot is designed to work next to a person; while the machine unloads a box or moves things around that need moving, the person can focus on tasks better suited for humans: prioritizing, making decisions, organizing, assembling.
The robot is supposed to be easy-to-use for a production-level factory worker. It doesn’t come with an IT/robotics geek or MIT PhD included, as Brooks puts it. It has to be self-contained. In fact the robot is notable for a number of things it doesn’t try to do. It doesn’t connect to a network or Wi-Fi (factories and plants don’t necessarily have connectivity). It doesn’t have speech recognition (factory floors are noisy) or a touchscreen (workers wear gloves). It doesn’t move around the room by itself. And it doesn’t require any software programming by its users. “Our whole thing is about lowering the barrier to entry,” Brooks says. “We’re not asking for anything except a 110 [volt] outlet.”
But it is designed to be a robotics platform. Brooks and his team are hoping to cultivate an ecosystem of robotic app developers. Imagine apps for different factory tasks, or eventually, all kinds of other things—more on that below. (Bartending, anyone? Wouldn’t a row of Baxters making drinks be better than waiting 10 minutes in a crowd every time you order a beer?)
Indeed, it’s tempting to call Baxter the “iPhone of robotics,” or some such. “I’d like to say it’s the Apple IIe,” Brooks jokes. “There’s a long way to go. But we’ve broken away from the mainframe, which is the current industrial robot.”
There’s no question this whole approach could take a while to catch on in industry. But here’s the plan: Rethink is going after small and medium-size manufacturers—say, businesses or divisions with 10 production workers—to be its initial customers. The first applications will be for materials handling. But subsequent software upgrades will enable the robot to do things like push a button, work with other machines to test or weigh parts, and use both its arms together to move a box out of the way, says Brooks.
“We’ve taken this robot to factories—never seen the factory before—and an hour after we pull up, it’s on the factory floor doing a useful task,” says Brooks. “We show it to the line workers, we show them how to program it, and in a few minutes they’re getting it to do simple tasks.”
Much has been made of Rethink’s big goal of revitalizing U.S. manufacturing. “It gives automation capabilities to small companies that couldn’t afford it before,” says Brooks. “They don’t have to put in safety cages, they don’t have to reconfigure. It’s not a long, drawn-out project, and it’s cost-effective to do a task for two hours. Whereas a traditional industrial robot, there’s so much overhead of getting that task set up. We think it’s going to make U.S. labor more competitive on low-end goods that currently we can only do in Vietnam. For a certain class of objects, and a certain class of tasks, it adds to the attractiveness of on-shoring, or at least not off-shoring.”
Interestingly, the robot has gone through six iterations in four years, says Brooks. The first prototypes had only one arm, and they were configured very differently from the current model. But going out to plants and talking to potential customers about the tasks that needed doing—as well as working closely with robotics suppliers and fabricators—led to the current design. And it is a fresh one: less than a dozen have been built so far, and some software debugging remains to be done. Brooks emphasizes that the robot can be manufactured cheaply because of advances in materials and many built-in design features.
One point of contention from entrenched vendors: “The conventional industrial robot manufacturers will say, ‘It’s not as precise as our robot.’ And you know what? It’s not!” says Brooks. “You move it 10 centimeters, it still works. You move theirs 1 millimeter, it doesn’t work. It’s not about precision, it’s about sensing.” Brooks continues, “A traditional industrial robot is measured by speed and repeatability. Ours should be measured by adaptability, flexibility, and ease of use.” But he admits: “When it’s a completely new category of product, is the world ready for it?”
I asked Brooks how this product launch stacks up against other famous commercial robots from his career, like Roomba and PackBot. “My aspirations are high for this robot. This is the first mass-produced, slightly sentient humanoid robot,” he says. “How well it sells, we don’t know yet. That’s why we’re VC-funded.”
He did offer this comparison to Roomba: “A big lesson we learned from the Roomba [was] they had to work out of the box. This is the first out-of-the-box humanoid,” he says. “These are the first ones going out in the wild.” (And coincidentally, he says, Rethink’s rollout comes exactly 10 years after Roomba’s release, down to the week.)
Brooks says the big challenge for Rethink now is refining its first product based on customer feedback—and figuring out what needs to be added to the robot over time, as well as eventually expanding the product roadmap. Some vestiges of his big-idea academic thinking come out as he talks about selling Baxter to researchers (for $22K they’ll buy it, he says) and providing a software development kit for them to experiment with the robot. “I think it’s going to be interesting that people will be able to use it,” he says. “I think they’ll try all sorts of different applications for it, ones we would never have thought of. It’s going to be really exciting to see which ones of those actually make economic sense.”
For Rethink, the near-term path is clear. “We have a solid economic case for manufacturing,” Brooks says. “But once we get it out there, who knows?”