Two years ago, Scott Eckert, while on vacation in the south of France, gathered his family around his laptop. The month before, he had accepted a job as CEO of a secretive start-up that was developing an industrial robot, and now he was about to see a video of the first demo of the machine.
He and his two children watched silently as the robot, which turned out to be no more than a small, cranelike arm, shakily grabbed and lifted a plastic disk. The video ended. His 6-year-old son broke the silence. "Dad, is that it?" he said. Eckert wondered the same.
Everything about the company Eckert would soon be running had been a bit mysterious. When the headhunter contacted him months before, he wouldn't tell Eckert much except that the company had been founded by famed scientist Rodney Brooks, who, until a few years earlier, had led MIT's computer-science and artificial-intelligence lab. Brooks, perhaps the most acclaimed roboticist alive, had co-founded iRobot, maker of the hugely successful Roomba vacuuming robot.
Eckert, a former Dell executive who co-founded Motion Computing, a tablet company in Cambridge, Massachusetts, met Brooks in a hotel lobby. At the meeting, Brooks continued to be sketchy on details, Eckert recalls, but he was enthusiastic. Brooks explained that he wanted to take robotics to the next stage by bringing out a game-changing robot for manufacturing companies. One that, compared with existing industrial robots, would be easier to deploy, more useful, and much, much cheaper--making it affordable even to small companies. Brooks said he was convinced his company could sell tens of thousands of them, even hundreds of thousands. Maybe millions. Brilliant young robotics engineers were already on his team and making great progress.
Eckert checked out Brooks every way he could and heard nothing but good things. He called many of his contacts in the manufacturing world. "They all told me, 'Hell, yes, we'd love to have a cheap, easy-to-set-up, capable robot; are you kidding?' " Eckert took the job, agreeing to start right after he whisked his family away for a long-planned vacation to France.
Now, having seen the video, he had to wonder: Had he made a mistake in putting his faith in Brooks?
But it turns out Brooks was just getting warmed up. That robot arm was merely the start. Brooks already had the team at his company, Rethink Robotics, hard at work building something much more ambitious. Called Baxter, it is a humanoid robot that has the potential to be everything Brooks was shooting for: a breeze to use, capable of handling any number of basic assembly-line jobs, and ridiculously cheap. Many experts would have said such a robot was a decade or more away.
In fact, Baxter is set to go on sale in October. But Brooks is the first to admit that the success of this product is no sure thing. For starters, it is at the absolute bleeding edge of one of the most daunting challenges facing scientists and engineers for centuries: how to endow machines with human capabilities. Getting Baxter to work has required innovating on multiple, complex fronts at once and then making it all work together seamlessly.
And even if Baxter works as hoped, there is no guarantee that companies will buy it in big numbers. "No one has ever done anything like this, so there's no way of knowing what companies will make of it," says Brooks. "Getting people to buy them is the one thing that we absolutely have to do. And it's the one thing that worries me."
Consider it the risk that goes along with a humongous upside. Brooks's plan for Baxter is so ambitious that it's almost scary: replacing humans in millions of jobs in the U.S. alone. Baxter can be taken out of the box, set up, trained, and put to work in about one hour. At $22,000 each--less than the price of a minivan--it could easily pay for itself in months, saving a company $30,000 a year or more in labor costs per robot. (This will be a good thing even for the workers who are replaced, Brooks argues, because they will get better jobs--but more on that later.)
That's just Baxter 1.0. A few years from now, Baxter could take over more complex jobs on manufacturing lines, such as operating machinery. Perhaps one day, it could also make a mark on service industries. Picture Baxter flipping burgers, tending cash registers, sorting files.
And Brooks hints at even more ambitious plans. He is aiming at no less than a revolution in how work gets done, one that would change the economics of labor. "This robot will just keep on improving, and doing more and more," he says.
I meet Baxter in July, at Rethink's offices in what the city of Boston has taken to calling its Innovation District, a once- and to some extent still-gritty area outside of downtown, on the wrong side of the charmless Fort Point Channel. The site on which the offices are situated happens to be the former location of the first electrified factory in Boston, once seen as heralding the future of manufacturing. It's a history hardly lost on the Rethink team. Inside, the offices are a cross between the usual high-tech swanky blandness and a busy workshop. The cubicles are punctuated with workbenches and shelving that are littered with all sorts of random-seeming odds and ends, such as cereal boxes and automotive parts. (These are items Baxter uses to practice on.) Also perched on some of the surfaces is the occasional humanoid robot or parts thereof.
Baxter is something like a cartoonist's vision of what the top half of a friendly, uncomplicated, but hardworking robot would look like. Robots can easily come off as either creepy or goofy, but Baxter strikes a pleasant enough balance between the humanoid and the machine, with a head consisting of a computer screen displaying a line drawing of a face. The particular robot in my demonstration is bolted to a pedestal behind a workbench, which would be a fairly typical arrangement at customer sites.
In this demo, the idea is that I will teach Baxter to move a random widget. I'll grab one of Baxter's arms, position it over a widget, and with the click of a button mounted on the arm, I'll confirm to Baxter that this is the object I want it to be able to recognize and pick up. Then, Baxter will pick up the widget in the fingerlike grippers at the end of one of its arms, at which point I will guide Baxter's arm to where I want it to place the widget. After that, Baxter should be able to do all that on its own, over and over again, as widgets parade by it on a hypothetical conveyor belt, unfazed by variation in how the widgets are positioned. Baxter should be able to stick the widget in a box, put it in an electrical tester and sort it into Pass and Fail piles, drive a screw into it--whatever's needed, as long as it's a fairly simple task.
Brooks has made it clear to me that this demo is premature--it's one of the first being given to the outside world, and Baxter is still undergoing intense tweaking. And sure enough, Baxter misbehaves at the start, having trouble with every step, finally becoming sullen, or so it feels to me, though technically it has merely become nonresponsive. An employee patiently reboots Baxter, as one would a balky laptop or cell phone. The second time around, everything goes pretty much as it's supposed to. Baxter works.
But not like other machines. In particular, Baxter doesn't--and wasn't intended to--move in the precise, sharp, angular way normally associated with robots (and with those tiresome people who imitate them). Instead, it seems a bit uncertain, taking its time and struggling a bit to get it right. As it does so, its screen-based face goes through a range of expressions, finally settling into one of contentment when it hits its groove. For an industrial robot, Baxter seems, well, laid back.
Brooks, on the other hand, hovers near the demo like an anxious parent. He is clearly trying to resist intervening but occasionally fails and jumps in to run things. He has participated in demos of various robots hundreds of times over the years, but he confesses he is nervous about this one. "We still have a number of bugs to work out," he says with a shrug. It's about two months before the official release date.
Later that day, Brooks regains his normal, affably edgy composure while he traces out his path from academic superstardom to high-profile entrepreneurship. When he stepped down from running the artificial-intelligence lab at MIT in 2007, he recalls, people were so shocked they asked him if he had a serious illness. What he had, in fact, was an urge to get back to the hands-on robotics tinkering of his early career. Brooks began toying with ideas for making robots more useful.
That, in a sense, is what all roboticists do. But Brooks has always had a more down-to-earth idea of what useful means. In the mid-1980s, when robotics was struggling to come up with complex software programs to mimic human intelligence, Brooks electrified the field with simple, insectlike crawling robots created with bare-bones programming. His machines excelled at real-world tasks, like getting around an office without bumping into things, and were relatively inexpensive.
Brooks became a clear leader in the field and drew all sorts of funding for experimental robots. But what he really wanted to do was see robots become integrated into everyday life, and that meant starting a business. His first was aimed at producing advanced toy robots that would cost just $100, leading him to tour Asian factories to understand the art of making things cheaply. The company he co-founded in 1990, iRobot, eventually ended up abandoning toys for robotic vacuum cleaners, scoring big with the Roomba's release in 2002. "When I started my career, there were a few hundred mobile robots in the world," he says. "iRobot made more than a million robots last year." The company's annual sales now top $465 million.
Brooks left iRobot in 2008, but the success gave him the first and perhaps most important of a series of what he calls maxims that guide him at Rethink, and with which he plies his employees in lieu of actually telling them what to do. ("I don't manage," he says. "I have no reports, and I don't want any.") That first maxim: Get something out that's so laden with breakthrough capabilities at such a low price that the competition will be crushed before it exists. "There are competitors to Roomba, but even today, they don't work as well as our first one did, and they cost more," he says.
Vacuuming proved a sweet spot in the previously nonexistent mass-robotics market, and Brooks was determined to find another. Having succeeded in the home, he turned his attention to robots for businesses. Though there wasn't much need for robots in offices, factories were another matter. Industrial robots have long been handling heavy-duty chores such as welding and spray painting for big automotive companies, but those can cost $100,000 or more. And then the real expense kicks in: They have to be carefully set up and programmed by experts to rapidly perform actions to precisions of one-hundredth of an inch, over and over again, without any variation. Assembly lines have to be built around them. And they have to be operated in caged areas away from workers to avoid crushing someone's skull in the blink of an eye.
For the most part, these robots are affordable only for the largest heavy-industry companies. What, wondered Brooks, would a robot for the rest of the manufacturing world look like and do? He pored over labor statistics on manufacturing. The new sweet spot, he decided, was materials handling, which basically means picking things up and putting them down, usually on assembly lines. A lot of those jobs could be handled by a robot with a dexterous arm. Narrowing things down by difficulty of task, he came up with a list of jobs that he was sure he could build a robot to handle. It was a list of jobs performed by 800,000 people in the U.S.
If Brooks could develop a $22,000 machine capable of grabbing parts and shoving them somewhere usefully, he was looking at a $16 billion market. By the summer of 2008, Brooks was raising money for his new robotics company.
Getting to Baxter was a journey. First, there was that lone, cranelike arm, an appealingly simple approach, and one that followed another of Brooks's maxims: Robots should not be humanoid just for the sake of making them look like humans. Doing so creates an aura of "inauthenticity" about the machine, he says, much like putting racing stripes on a Toyota Yaris. But the crane arm didn't seem very exciting. Besides, there were already companies making $30,000-and-up robotic arms that could be programmed to carry out simple tasks, and Brooks wasn't gunning for mere incremental improvement. "People work better with two arms," he says. "Why wouldn't a robot?" (He briefly considered three arms before rejecting it as too much of a good thing.)
Having two arms meant mounting them on a torsolike frame. The middle of that frame seemed a natural place to mount the screen that would facilitate interacting with the robot, but the team found that the moving arms often blocked the view of the screen. So the screen was mounted on top of the frame, making it look like a head. Uh-oh; with two arms, a torso, and a head, Baxter was suddenly looking distinctly humanoid. "But it was authentic," insists Brooks. "Every feature had a purpose."
Those features eventually came to include a face, or part of one. Baxter is designed to work right alongside ordinary employees. That meant there had to be a way for Baxter to let those standing near it know what it was up to, lest they become startled or get in the way. Originally, that chore was left to a series of lights on top of Baxter's "head." Though the lights are still there, they were deemed insufficient, so Baxter got cartoon eyes on its screen. Its eyes "look" in particular directions to indicate where an arm is about to reach. A calm, focused expression indicates Baxter is successfully working away; the eyes open wide in surprise when someone suddenly approaches; and they are canted in confusion when Baxter is not able to complete a task.
From the beginning, the technical challenges were fierce. Robotics labs today are filled with million-dollar, state-of-the-art robots that struggle with mundane chores, such as gripping a cup of coffee or opening a door, even after months of programming. Baxter, on the other hand, would have to perform any of a range of tasks with whatever object was placed in front of it, and it would have to learn one of those tasks in just minutes. Every aspect of its target capabilities--recognizing objects, gripping them, manipulating them, being easily trainable, not being dangerous to proximate humans--was in its own right a massively complex goal that was pushing the envelope of robotics. To work on them all at once and tie them all together would have been a frighteningly daunting challenge if Baxter was planned as a long-term project and a half-million-dollar machine. To do it in a few years for $22,000 seemed at times just plain impossible.
That's where Brooks's nonmanaging management style came in. Development-team members are quick to cite his role in keeping everyone moving forward in the face of an endless stream of technical impasses and sagging hope. He dispensed encouragement, ideas, and reminders of the vision. "He's pretty selective about when to get involved," says Ben Berkowitz, one of the lead engineers. "But when he tells us something, it's usually important, and it's great."
Robotics engineers typically make something work any way they can and then worry later about how much it all costs. But from the beginning, Brooks forced the team to focus on low-cost solutions. He hired a vice president of manufacturing before there was even a basic design sketch. "We were able to put a ton of effort early on into making this buildable," says Jim Daly, the fellow who took the job.
With Daly's input, the team used a lot of plastic, off-the-shelf parts for the components. Brooks was determined to outsource all manufacturing. Rethink wouldn't make a thing. Instead, different companies would make pieces of Baxter--the arms here, the electronic controls there, and so forth, with final assembly handled by yet another company. The result, says Daly, is that Baxter will be profitable at fairly low volume. He won't say how low, but he notes that volumes won't have to get to nearly the levels the Roomba required to pay off. "We don't need to sell millions," he says. "If we can crank up volume just a bit, we'll be good. Baxter is no loss leader."
Brooks was as uncompromising on safety as he was on cost. One errant swipe of a robotic arm at a customer site as a flimsy human happened to be walking by could bury the venture and set robotics back a decade. Baxter fairly bristles with features to prevent that sort of inadvertent aggression. Besides the facial expressions that serve as warnings, Baxter features several cameras and sonar sensors that monitor the surrounding area, so it "knows" when someone is nearby and freezes when there's a risk of impact. If someone manages to shove a body part in front of Baxter's moving arm too quickly to give the robot time to react, the pressure of the resulting impact on the robot's yielding plastic "skin" triggers an instant halt to all motion, with the result that Baxter can't deliver much more than a love tap. (The company has a video of Eckert being struck repeatedly on the head by Baxter to no ill effect.)
The most ambitious element of Brooks's scheme was designing Baxter to be trained the way humans learn things--by having someone show them how to do it--instead of having to be programmed by experts. At first, the team came up with a design that would let someone train Baxter on a new task by guiding it via joystick. But Brooks rejected the approach as nonintuitive, demanding instead that Baxter's arms be directly guidable by hand, the way a parent might help a child master the use of a knife and fork. Building guidable arms turns out to be a lot harder than it sounds, requiring the design of a "zero-gravity" system that allows Baxter's arm to move freely when being pushed around but prevents the arm from flopping under the pull of gravity. The team was skeptical of the idea up until the moment it finally worked. "Once they tried it, they instantly got why it was so much better than a joystick," says Brooks.
After being trained on a task, Baxter goes about it much as a human would. That's because Baxter isn't designed to maneuver its grippers in the exact same way every time, as a conventional robot would. Rather, it spots the object and then "feels" its way through gripping and maneuvering it, doing so a little differently each time. If asked to stick a part into a tight-fitting box, for example, Baxter carefully wiggles it in, sensing the part slipping into place--which is how you would do it, too. This ability to use vision and touch sensing to improvise its way to successful task completion makes Baxter highly adaptable to the ordinary, human-oriented work conditions found in small-company assembly lines. The idea is that Baxter can be wheeled right in to do what a person was doing, rather than needing a work environment to be built around it, as with conventional industrial robots.
The team is still struggling frantically to smooth out the sorts of bugs and rough edges in Baxter that appeared early in my demo. But all in all, the robot seems to be coming together, much the way Brooks envisioned it. Which merely leaves the nagging question of whether companies will actually buy a product that may be perceived by some as alien and baffling. "After six months of production, we will have doubled the world's population of humanoid robots," he says. "Is it too big a step for the market?"
When approached by Rethink, potential buyers all ask the same question: What's the robot's precision and repeatability? That's how conventional industrial robots are measured. "It's the wrong question for Baxter," says Brooks. "They might as well be asking how high it hops."
Clearly, Rethink has its work cut out for it when it comes to convincing manufacturing companies that Baxter's imprecise, improvisational approach is an asset, not a handicap. To make things trickier, the company will be trying to sell to the small- and midsize-business market, one that is largely unfamiliar with robots. "Our main target will be the 107,000 manufacturing companies with between 10 and 500 production employees," says Eckert. "But we'll also be selling to the 160,000 manufacturing companies that have fewer than 10 employees."
But the pitch has its appeal. Because conventional robots are so difficult to program and set up, they can't be cost effective unless they remain in the exact same place doing the exact same thing for years. Baxter, in contrast, could work much like a temp. With 10 minutes of training from any employee, it can get to work on a job and after a few hours switch to another one. It could relieve different workers at different times, and roll with changes in the assembly line.
That's why Chris Budnick, who has had a chance to try Baxter out, wants to buy one. Budnick is president of Vanguard Plastics in Southington, Connecticut, a 30-person injection-molding firm that makes parts on contract for medical-equipment makers, automotive companies, and other manufacturers. Vanguard already uses more-conventional robot arms to yank parts out of its molds. The arms are well suited to that job, he says, because it takes precision, it doesn't vary, and the arms' high speed keeps the expensive molding machines operating at peak output. But then he has to pay people to stuff the parts into bags and boxes for shipping--using robot arms would require professional reprogramming for each new part or package. Rethink, looking for user feedback, let Vanguard put Baxter to work earlier this year. "We had him picking and packing in five minutes," says Budnick. "We think it would pay for itself in a year. We'd have it working three shifts and packing two different lines at once, one with each arm. The sky is the limit with this thing."
It's not just that Baxter could save labor costs by replacing a human employee. It could also open up opportunities to reimagine the manufacturing process. Baxter won't object to working in less-populous locations, where it's cheap to build but hard to find labor. It won't get sick from working in extreme heat or cold, or from exposure to toxic substances or poor air quality. It will uncomplainingly work graveyard shifts, weekends, and holidays. It won't get repetitive-stress injuries from repeating the same motion 10,000 times a day.
Working against Rethink will be the inevitable concern that the company is facilitating the outsourcing of jobs to robots. Automation has been replacing people in the workplace for two centuries, of course, but there's something particularly unsettling about a humanoid machine being dropped in as a one-to-one substitute for a flesh-and-blood employee.
The last thing we should be worrying about is protecting the human's place on the assembly line, counters Brooks, who has visited numerous factories in many countries. "These are mindless, repetitive jobs that turn people into robots," he says. Brooks adds that many factories are increasingly having trouble finding people willing to fill them, especially in late-night and weekend shifts. He reasons that if robots take those jobs, the cost savings and productivity boost will create growth, opening up positions more interesting than assembly-line work. As a result, says Brooks, employees replaced by robots won't be let go; they will be promoted. "The PC didn't replace people in offices," he says. "It made them more productive."
Let's hope Brooks is right, because if Baxter's capabilities continue to improve as envisioned, it could take away tens of millions of jobs in the manufacturing and service industries. The resulting productivity boom could lead to an economic renaissance--if it doesn't merely lead to vast layoffs.
But first, Baxter has to work, and well enough to get good reviews from early customers. So, Rethink is being careful about who gets the first models. "We want them to go to companies that will put them in the sorts of applications we know they can succeed at," says Brooks--tasks such as picking up an object from a conveyor and putting it into a bin.
Given that Amazon founder and warehouse-automation pioneer Jeff Bezos was the first to invest in Rethink, there's probably more to Baxter's future than picking things up and putting them down. In particular, says Brooks, Baxter's abilities could grow quickly after programmers get their hands on its software innards. And they will soon enough, because Rethink will encourage outside developers to take Baxter in new directions. Baxter may well become the iPhone of robots, a popular platform backed by an array of third-party apps. "At some point, we expect one of our manufacturing partners will find the right application for Baxter," says Daly, "so he can help build himself."
Still, Brooks expects many--maybe most--businesses to initially dismiss the idea that they are ready for robots. He says he got much the same reaction when he started talking about robot vacuum cleaners. Brooks took to asking the skeptics if they could imagine robots vacuuming homes in 50 years. "They always said they could," he says. "At which point I told them that that meant we could agree that sometime between now and 50 years from now, robot vacuum cleaners would be coming into homes. All we were doing is arguing about where along that timeline the change would actually happen."
For Roomba, it turned out to be a couple of years. Now, the clock has started ticking on Baxter. "I think everyone's going to be surprised," says Brooks.