A low-cost, easy-to-train robot that's safe to work with people is designed to make American manufacturers more competitive with low-cost regions

Mark Shortt
Editorial Director
Design-2-Part Magazine

Momentum for reshoring had been steadily gaining steam. But when Rethink Robotics unveiled its Baxter robot last fall, it provided a powerful new boost to efforts aimed at bringing manufacturing back to America. Baxter, described by Rethink as "the world's first humanoid robot capable of applying common sense behavior to manufacturing environments," is a mobile, 165-pound American-made robot—with a price tag under $30,000 and an array of user-friendly features—that's designed to make U.S. manufacturers more competitive against regions that employ low-cost labor.

"Companies are just starting to use the robot; it's just beginning to make it to manufacturing lines right now," said Mitch Rosenberg, vice president of marketing and product management for Rethink Robotics, a Boston-based startup backed by Bezos Expeditions and the venture capital firms Charles River Ventures, Highland Capital Partners, Sigma Partners, and Draper Fisher Jurvetson. "But I would say the universal reaction is 'We didn't think this was possible.' That's the most frequent comment I hear. The universal reaction is that they're really thrilled to have them, they're having a lot of fun, and they didn't think this was possible."

Baxter is unlike traditional industrial robots in a number of ways, beginning with an unusually low price point that puts it well within reach of small and mid-size manufacturers. Designed to work safely around people  in a production setting without safety caging, Baxter can perform a wide range of simple, repetitive tasks—including part and material handling, machine tending, testing and sorting, and light assembly—that have traditionally been lost to overseas labor. Besides having an unusually low base price of $22,000, Baxter takes affordability to a new level by eliminating the combined expenses of caging, integrating, and programming a traditional industrial robot.

"Manufacturers tend to be risk averse, and this price point makes it much more attractive for them to take a flier on a new technology," said Rosenberg. "So many of them are saying, 'Because of this price point, this doesn't feel risky at all.' That's a big factor."

Baxter is based on the vision of Rethink's founder and chief technical officer, Rodney Brooks, the co-founder of iRobot and former director of the MIT Computer Science and Intelligence Lab. By breaking down the cost and safety barriers that have long limited the use of automation in American manufacturing, Baxter is said to "redefine how small, mid-size, and large domestic manufacturers use automation to compete with manufacturers in low-cost regions of the world." Its design is centered around the idea of making what Rethink's vice president of product development, Elaine Chen, calls "an incredibly inexpensive robot" that balances cost with performance (see Building Baxter in the U.S.)

"We embraced Design for Manufacturing and low-cost manufacturing techniques very, very early on in the design cycle," Chen said in a phone interview. "Consequently, all of the parts that we have in the robot are mass producible using cost-effective manufacturing techniques, such as die casting and injection molding. We don't have any high-cost manufacturing processes of any significance in the robot. We've designed for a low-cost bill of materials and also low-cost manufacturing techniques."

From a hardware standpoint, Chen said that Baxter is different from other industrial robots in that it's a low- payload robot that's very safe to be working side-by-side with people. "We've got load-level algorithms to detect forces at every joint," she noted. "If somebody actually gets in the way of the robot arm while it's moving, the robot arm would detect that and stop immediately."

Chen often demonstrates this capability in her tradeshow demos. "I'll let it get going, and then I'll kind of stick my head in the way of the elbow, and it will just stop immediately; it wouldn't hurt. So that's pretty innovative. We don't know anybody else whose robot has such a sensitive force sensing threshold for people detection."

Brooks, in a statement accompanying Baxter's launch, noted that roboticists have had previous successes in designing robots that can operate with super-human speed and precision. "What's proven more difficult is inventing robots that can act as we do—in other words, that are able to inherently understand and adapt to their environments," he said. "We believed that if we could cross that chasm with the manufacturing environment specifically in mind, we could offer new hope to the millions of American manufacturers who are looking for innovative ways to compete in our global economy."

One of Baxter's unique capabilities is its ability to apply common sense to its environment. If Baxter drops an object, for example, it knows to get another before trying to finish the task.

"'Common sense' means that the robot does what you expect, and signals to you what it's going to do, so it interacts with you in a way that's already familiar," said Rosenberg. "You don't have to learn how to interact with this robot. If you, for example, are asking the robot to pick up items off of a conveyor belt that's running a certain speed, and then you increase the speed of the conveyor by 20 percent, with a traditional robot, you'd have to reprogram for the new speed. You'd have to explain to the robot—using software and signals—that it had changed. With our system, just like with another human being, the robot automatically detects that things are going faster. Moreover, it automatically changes the way it guides its arm to pick up that item on the conveyor, exactly as your common sense would tell you a person would do it. That's what we mean by common sense: It means that you don't have to think of every detail that the robot has to know; the robot automatically adapts its behavior the way a human would."

A driving force behind the development of Baxter has been to distinguish it from previous robots that were all about speed and precision, by building, instead, a new type of robot that's more about the ability to sense and adapt to its environment.

"What that means is that our robot can work in a very unstructured environment," said Chen. "You don't have to always present the part in the same place. So, just as a human might not actually present a part to another human always in the same spot, the robot can actually use its vision to find the part and pick it up, even if it's a couple of inches to the left of where it showed up the last time."

Along with various sensors, sonar, and cameras, software is crucial to the development of Baxter's unique capabilities. Rethink's product strategy involves introducing a solid hardware platform, and then continually adding value to it by releasing frequent software updates. The company is currently working on a software development kit (SDK) for the manufacturing robot, but it has not been released for public use. In April, Rethink released its first feature update for its research robot, enabling users to customize the robot for specific types of work.

"With a very simple procedure, users can download the updater, update the software on the robot, and get a whole range of new functionality that enables them to do much more complex tasks," said Chen. "That's our strategy moving forward. We will have a steady cadence of new capabilities that bring some benefits to customers through regular software updates."

If this sounds familiar, there's good reason: It's exactly what the PC industry pioneered, and is now a fact of life with cell phones and other mobile devices. Rethink's innovation is to apply the strategy to an industrial product.

"We believe that that's really new and particularly valuable for our industrial product," said Rosenberg. "One of our motivations for offering the Baxter research robot is the idea that if researchers—both in industry and in academia—come up with nifty new applications for the robot, or nifty new ways to do things, we'd like there to be an ecosystem for taking advantage of that software. They can sell it on the open market or let us license it back from them, so that just like PCs and cell phones, the devices become more and more valuable over time, rather than depreciated and less valuable over time, the way traditional hardware based products are."

Rosenberg said that interest in Baxter is such that "plenty of corporate researchers" are buying the robot and using it, even without the ability to write unique application code for it. "It's really excited the imaginations of folks in both the manufacturing lines and the manufacturing labs," he said. But Rethink's strategy of opening up the development of the software to users reflects its growing recognition that the sky is the limit for future applications.

"We've chosen to focus our efforts on making a robot product that will help small and medium size businesses succeed in bringing manufacturing back to America," said Chen. "That's our mission. But we also realize, in creating the hardware platform, that this is a very general-purpose humanoid robot that can do a lot of other things. We have a lot of inquiries from people that are interested in using this robot for other verticals, like elder care and assisted living. So we thought the research SDK would be a good way for other people to explore those verticals."

Manufacturing environments, in the meantime, are loaded with possibilities, including pick and place tasks, light assembly, material handling, machine tending, testing and sorting, and—eventually—finishing operations. "We're seeing a number of, not just customers, but increased interest around packaging," said Sue Sokoloski, Rethink's director of marketing, at a product demo on the floor of the ATX East Automation Technology Expo in June. "Packaging seems to be a point of previously un-automated jobs. We're seeing a lot of interest in the 'end of the production line' type of jobs."

"They want parts handled, but you don't need a computer science person to reprogram the robot on the line every few days," said Anthony Robson, Rethink's product manager, also at ATX East. "That's Baxter's sweet spot right now."

"Baxter is great at pick and place," added Rosenberg. "It can pick from one surface and put it in another; it can pick out of a bag and put it on a conveyor or vice versa. We just recently released software to the marketplace that allows Baxter to interface to other machines. There are plenty of machines, obviously, in the manufacturing environment that can exchange signals so that they can sequence operations in an intelligent way, accept inputs from other sensors, et cetera. So that's another set of applications that we're pursuing.

"We're also very excited about the idea of using Baxter to tend other machines," he added. "You can imagine that there are human beings that do repetitive motion every day, just pushing the button on a polishing machine, for example, to make that machine go on. That's an example of machine tending that requires some level of intelligence that is really not worthy of having a human being perform it, and it's something Baxter could be very perfectly suited to. Eventually, we could imagine Baxter doing more physically demanding tasks, like finishing operations, like polishing or sanding surfaces, but that's not something that we're releasing right now. The simple summary is that there are a wide variety of manufacturing tasks that aren't necessarily assembly, but require a lot of time and a lot of money and a lot of footprint in the factory, that Baxter could do very, very well."

Many of these processes have been difficult, if not impossible, to cost-effectively automate in the past. Not only do traditional robots themselves carry price tags often exceeding six figures, but, unlike Baxter, they need to be programmed—a task that can take several months. After coming up with an idea to automate a production line, it could take up to nine months and several hundred thousand dollars before a traditional robot is actually ready to do something productive on that line, according to Chen.

"That only works when you have a product line that's got a stable enough product going through it that it's going to go long enough to make it worthwhile," she said. "The product also has to be high value enough to have the market to support that. Whereas, our robot is different in that basically, you do not program it; there's no integration necessary.

"Our vision is very much around having machine operators working on the manufacturing floor be able to go up to the robot and grab one of its hands and show it what it needs to do by demonstration," she continued. "So how we train the robot to do that is entirely an integrated user experience—using the screen on the head, and the buttons, cups, and knobs that are on the arm and on the torso. There's no programming at all in the manufacturing use case, and anybody can go up and teach the robot how to do something new in a matter of minutes to a maximum of a couple of hours. So then suddenly, some of the work that never would have been cost-effective to automate using the traditional way—something like a short run project with only 10,000 parts, for example, and you're due to be done with that project after two weeks—could be automated using this robot."

Many small and medium-size manufacturers, which often have less capital and do more of the short run operations that Chen described, are taking notice. Sokoloski said that the company doesn't disclose the number of manufacturers that are currently using Baxter, but that it's currently "in the hundreds." Baxter's affordable price tag and the fact that it doesn't require expensive programming are enabling many of these companies to automate for the first time and, as a result, compete directly with overseas manufacturers.

Rethink is not only making a robot designed to keep production from migrating overseas; it's also "putting its manufacturing where its mouth is" by building Baxter in the United States. More than 75 percent of the value of the robot is made in America.

"It starts with our original vision for the purpose of the company," said Rosenberg. "Our founder, Rod Brooks, felt that there was no reason at all why America should lose its manufacturing base given our level of skill and innovation, and his idea was to create an innovation that made it possible for manufacturing to take place economically in the United States. You may be familiar with the term 'reshoring.' The idea that we would be actively advocating this kind of change in the American economy while simultaneously contributing to the opposite trend didn't seem to have integrity. So that was the first reason why we decided it was important to be made in America.

"As the marketing guy, I can tell you that this is a universally applauded attribute of the product from the point of view of customers and distributors," he added. "There is so much enthusiasm in this country for being a serious player in manufacturing again, and they love the idea that they can, in fact, do that while automating with an American-made robot."

"From a technical standpoint, I can tell you there are extreme benefits to keeping innovation very close to manufacturing," Chen added. "Our contract manufacturer is within driving distance of headquarters, and that has made the process of actually bringing up the manufacturing line so much better. We're able to go up there every other day if necessary. If any issues come up, we can go there immediately. Having manufacturing close to innovation is really, really key to our successful journey in bringing up a manufacturing line and producing a quality product within a very short period of time. I don't think we could have possibly done that, even if it was in Mexico. I don't think we could have done that."

Another reason for making the robot in America, Rosenberg added, is that Baxter represents "a huge amount" of intellectual property. "Many elements of it have never really been done before, and we want to protect that intellectual property by making sure it doesn't get copied and knocked off and manufactured by others," he said. "So there are internal strategic reasons, in addition to market reasons, technical speed to market reasons, and general positive feelings about Made in America. There are reasons that are across the spectrum of business purposes [for making the robot in America]."

What type of feedback has Rethink been getting from companies that have used the robot so far?

Anthony Robson, product manager for Rethink, spoke with Design2Part Magazine on the floor of the ATX East Automation Technology Expo in June. He said that feedback on Baxter has been phenomenal. "They love it—the fact that in a couple of hours, they can be Baxter experts and understand how to use it in different ways," Robson said. "I think Baxter resonates with people in the manufacturing environment. I've had people say, 'I don't even know if I have an application,' and they've bought one anyway. They buy into the mission of Rethink Robotics and are willing to buy one on spec, just in anticipation that they'll find a way to use it because they identify with what we're trying to do."

"I think the most telling is that we'll ship out a Baxter and we'll say that we'll be on site the next day to set it up with you, and before you get there, it's like a kid on Christmas," said Sokoloski. "They've already unpacked it, put it on its pedestal, figured it out, and they're already getting it to work well before we show up. I think that's the greatest testament to the robot. They've already figured it out, and we haven't even landed on site yet. It's pretty spectacular."

Building Baxter in the U.S. Jim Daly, vice president of manufacturing and operations at Rethink Robotics, has built an enviable track record of launching and ramping innovative new products with previous companies. He's worked in the U.S., Mexico, and Asia, contributing his talents to capital equipment companies, contract manufacturers, and virtual operations at startups. Previously vice president of operations at consumer startups Zeemote and Tea Forte, Daly has also worked as director of manufacturing operations for Handspring, the maker of handheld products that was acquired by Palm, and as manufacturing director of Solectron's Complex Systems division. "Early in my career, I set up robotic assembly cells for semiconductor modules, so I've kind of come full circle," he said in a telephone interview. "I spent several years at the contract manufacturer Solectron, out in California, setting up operations to build their systems integration business. At Solectron, we'd get to work with many leading product companies, so the great thing there was to be able to introduce their products and see all of their innovation on a quite frequent basis." Daly's experience also includes work on setting up a factory for Apple to build Macs "when they were unable to keep up with demand," he said, as well as building and introducing several inkjet printer products for HP, both in the U.S. and Mexico. While at Juniper Networks, the maker of videoconferencing products, medical equipment, and semiconductor test equipment, the facility where he worked won the Industry Week Best Plant Award and Malcolm Baldrige award for the second time, he said. "It was a real good opportunity to work with many, many innovative product companies," he said. The challenge for Daly these days is to apply his manufacturing smarts to the making of a new type of robot that's specifically designed to help American manufacturers become more competitive—one that's surprisingly affordable, safe to work alongside, easy to train, and capable of sensing and adapting to its environment while performing the types of tasks that have long been consigned to low-cost overseas labor. In a recent phone interview, we spoke with Daly about the design and making of Rethink Robotics' Baxter robot.  Following is an edited transcript of our conversation.   D2P: How do you see this opportunity to introduce an innovative robot, such as Baxter, as different from any of your other previous projects? Jim Daly: This is certainly a very complex product. But as a manufacturing guy, one thing that's really cool is to be working at a company that is solely focused on really trying to bring manufacturing back to the United States. Baxter, as a product, is innovative in itself, and it's been a creative project to design it and launch it and build the operation, but equally, it's just a great story for how we're able to help our suppliers and partners and customers become more productive by bringing the robot forth.   D2P: What would you say is innovative about the robot? JD: Probably the first thing is that it's collaborative and safe to work around people. If you look at a traditional industrial robot, it's behind a cage, and people can't interface and be collaborative with it. But you can walk right up to Baxter, it knows you're there, and it's safe to be on a line and working side by side with people in manufacturing. Also, its user interface is incredibly intuitive. You don't actually program the robot; you teach it. An ordinary user can come up to the robot and, within minutes, really, train it to do a simple task. For complex tasks, if you're interfacing with other pieces of equipment, like conveyors or PLCs, it might take a few days to have that up online, compared to many, many months for a complex industrial robot to be integrated into a line. The other thing that's very innovative is its cost: At $22,000, it's incredibly low cost for the equipment that you're getting.   D2P: What were some of the design challenges that you encountered in making this new type of robot? JD: As we talk about cost, it's a balance of cost and performance that has been a real major driver of many of our decisions. We obviously have to be able to deliver economic value in how the product can perform, but by keeping that cost low, we needed to make trade-offs in materials and processes, making the robot easy to train and safe to be around people. Again, some of the decisions we had to make to allow for user interface, for safety—not just in the robot's operation, but in the materials that we're using. We need to keep the weight low because we want this robot to be mobile and able to be repositioned in the factory for multiple operations.  Keeping the weight low—basically throughout the robot, but particularly in the arms—is important.   D2P: How did your design goals for cost and quality influence the selection of manufacturing processes and materials? JD: We had the need to balance cost and performance, and the way that we decided to do that was to really develop a highly leveraged supply chain infrastructure and processes to be able to get the cost and scale to what we needed to achieve. One of the ways we did that was through tooling. We're more of an enterprise scale business than a consumer scale business; however, we took some of the processes from the typical consumer type environment and applied them to our business. We purchased roughly 122 production tools to be able to achieve the cost and scale that we wanted to be able to do. We'll never actually leverage the volume that we'll need off of those tools that allows us to be really, really competitive and get the quality and the cost down for those components. So, early on, we decided that we wanted to design this product for manufacturing, and DFM became a big part of what we did. When I came on board, we set out some principles. One, we wanted to modularize the product design so that the product had to be built in modules that we could potentially job out to different experts in the field. That could keep our costs low. We also wanted to be able to make sure those modules are fully tested before they move on to the next step in the process, to be able to keep the quality high before value is added. We brought suppliers into the design process early on. Before we even designed the product, our manufacturing partners with key processes were brought in to help influence the design. Then we said, 'We want to build in the U.S.,' and so we wanted to find partners that were close to home, that could be incredibly competitive to allow us to do it here, and then be able to do it quickly. So time to market was also an influence. Our suppliers had to be able to be innovative and progressive. They had to be cost competitive, so they had to be very competitive in how they were efficiently managing operation and scale. But equally important is a shared culture and values. This is a startup, and we're pushing the envelope. We have a culture of innovation here, a culture of openness, and we wanted to find partners that would share that. So we have very open relationships, very collaborative relationships with most of these suppliers. Somebody that would be looking to do a transactional based partnership or sourcing activity wouldn't fit with what we're doing here.   D2P: Did the fact that your suppliers were located close to your home base make it easier for you to innovate at any point in the process? JD: You bet. As I said, we brought them on early. I'll give you a couple of examples. Our gears are made of powdered metal, and to be able to get the quality of gear we want, we didn't want the cost of a turned gear, a ground gear. We needed to get sintering, but we needed to help them improve their processes enough to be able to get the quality that we wanted. I brought the guys from our chosen powdered metal partner here and sat them down with our design engineers. It was actually a pretty interesting process because the gear guys had been doing this for years and years, and our brilliant engineers said, 'We don't really need the help.' And this was a prototypical DFM early supplier involvement. We were able to get the industry knowledge from the suppliers and help influence our design, so ultimately, we ended up with a design that was manufacturable and we were able to push the envelope of the supplier's process to get a higher quality gear out of them. So we took that and we extended it other processes; die casting is another example. We have housings that go on the gear boxes, and we said 'OK, let's bring the die cast guys in. By now, our engineers are open to learning from the process experts, and so we brought the die cast guys in, and we talked about how to design the product and how to design drafts and flows. Typically, when you're doing a tool design, you design the tool, you try the part, you throw that part over to the tool designer, and they throw it back and say, 'Make these changes.' We were able to cut almost all of that out. So after the influence that they had on our design activity, we were able to send designs over and, in many cases, the tool designer came back and said, 'No changes required; everything is perfect.' And we were able to cut weeks and, sometimes, months out of the tooling process.   D2P: What are some of the major manufacturing operations that are used to make the Baxter robot? JD: There's a significant amount of injection molding.  As I mentioned, we had roughly 122  production tools, and injection molding makes up a big chunk of those. We also use thermoforming, die casting, sintering, powdered metal, extruded aluminum, tube welding, and some turning.   D2P: Approximately how many parts go into the making of Baxter? JD: Baxter's assembly BOM is made up of approximately 530 unique part numbers.    D2P: Can you talk a little bit more about the need to reduce weight, especially in the arm? JD: If you think about it, if the robot has to lift its arm, and if that's too heavy, then that reduces the payload capability. Even an end effector that we use on the end of the arm has to be light because the end effector itself is a weight. So when we were designing the robot, we went through and looked for opportunities to reduce weight.   As a matter of fact, the DFM initiative that we did identified redesigning the gear boxes in plastic at the end of the arm. If you think about your wrist, your shoulder is much stronger than your wrist, and so the upper joints—the shoulders and elbows—are all die-cast aluminum housings and sintered gears. As you move down the arm towards the wrist, the gearboxes are smaller and have less torque on them. Working with one of our injection molding suppliers who does very tight precision injection molding, we came up with a glass-filled plastic that provided structural performance similar to the die cast aluminum at significantly lower weight and cost. It also allowed us the added benefit of combining several parts. So, whereas with die cast aluminum, we would have to have three or four different parts to make up the geometry, we were able to cast it in one or two parts, or mold it in one or two parts, which allowed us to cut down on assembly labor as well.   D2P: What keeps you up at night, more than any other issue, when thinking about Baxter? JD: I've got to tell you, we've built an incredible team here, in the operations and also the extended virtual operation's partners that we have. And so I don't stay up at night worrying about us being able to build a quality, low cost product. If I stay up at night, it's probably more to do with thinking about 'Where can we use Baxter within our one of our suppliers, or within our contract manufacturer, to practice what we preach even more, and become more competitive ourselves. These days, as I go through our factories, I'm racking my brains on 'Where could we use our own technology?'   D2P: What possible types of improvements might we see in the next generations of Baxter? JD: The first thing is that Baxter is a platform, so, like a PC or your smartphone, it's continuously upgraded. Hardware-wise, it's stable. The plan is to continuously upgrade with regular software releases—we recently released the ability to integrate with PLCs—and also supplement the product with different end effectors. As we introduce more product to customers and we see what they're doing, we look for opportunities to maybe come up with standard end effectors that can be used on multiple products. We're also working on a non-vertical pick and place, and we're always working on improving the speed and accuracy of the product, so you'll see more of that.   D2P: I see from your website that Baxter doesn't eat pizza. What are the chances of Baxter becoming a little more nutritionally conscious in the future? JD: [Laughs] Boy, I don't know about that. I think we talked about Baxter being a barista, so maybe we can make lattes in the future.

This technical information has been contributed by
Rethink Robotics