JACOB GOLDSTEIN, HOST:
Inside the Defense Department, there is this agency called DARPA. Maybe you've heard of it. It's pretty famous - at least it's, like, nerd famous - for being way out there on the technological frontier. DARPA helped created the internet and GPS, recently funded work on, and I'm quoting here, "remote control of brain activity."
NOEL KING, HOST:
Another recent DARPA project - a sewing robot.
STEVE DICKERSON: We were the odd man out (laughter).
GOLDSTEIN: This is Steve Dickerson, the sewing robot guy. He's an emeritus professor at Georgia Tech. And in 2011, he responded to a DARPA call for ideas about manufacturing.
DICKERSON: No one at DARPA thought they were going to get a proposal for automatic sewing (laughter). You know, they didn't have a - they didn't have a clue.
GOLDSTEIN: Still, in his proposal, Dickerson pointed out that there is a military connection.
DICKERSON: The military must, by law, make all of their sewn products in the United States.
GOLDSTEIN: Like uniforms?
DICKERSON: Like uniforms.
KING: That law goes back to the 1940s. And at the time, there was a giant apparel industry in this country. So back then, when the Army and the Navy wanted to buy uniforms, they had lots of U.S. companies to pick from.
GOLDSTEIN: Over the past few decades, though, almost all the factories making clothes in the U.S. shut down. Today, about 97 percent of our clothes are imported. They're made by tens of millions of people who are working largely in very poor countries. So when Dickerson's sewing robot application came in, DARPA said, all right, here's a million dollars. See what you can do.
(SOUNDBITE OF DAVID ISAAC FELDSTEIN AND MARC FERRARI SONG, "THE BEST AROUND")
GOLDSTEIN: Hello, and welcome to PLANET MONEY. I'm Jacob Goldstein.
KING: And I'm Noel King. Today on the show, the surprisingly difficult quest to build a sewing robot.
GOLDSTEIN: Steve Dickerson wants to bring clothing manufacturing back to the U.S. If he succeeds, he's going to go way beyond uniforms.
(SOUNDBITE OF DAVID ISAAC FELDSTEIN AND MARC FERRARI SONG, "THE BEST AROUND")
KING: A couple of years ago, PLANET MONEY did this project on how a T-shirt is made.
GOLDSTEIN: Yeah, we followed it all around the world, right?
KING: From, like, the cotton to the...
GOLDSTEIN: To the shirt.
KING: The shirt, right.
GOLDSTEIN: And, you know, it's incredible how much of the process of making a T-shirt is, like, crazy high-tech.
KING: Right. There is the moment you pick the cotton, which is done by these big self-driving machines, and then you ship the cotton off to a factory, and it's these computer-automated machines with crazy names that make it into yarn.
GOLDSTEIN: But the next step, where the fabric actually gets sewn together to make a T-shirt, that is really low-tech. That looks like it could be 100 years ago.
KING: Yeah. It's real people. It's mostly women, sitting at sewing machines, feeding fabric in, sewing pieces of fabric together.
GOLDSTEIN: There used to be factories in the U.S. where people did this. And, in fact, Steve Dickerson, the guy who got the DARPA grant for the sewing robot, he saw them up close. He has family in Commerce, Ga. - that's this little town - and like a lot of little towns in the South, there used to be factories there full of people sewing clothes by hand. But at some point, it stopped making economic sense to do this kind of work in this country. You know, it's labor-intensive, it's low-tech. So the sewing industry moved offshore. Those factories in Commerce, they shut down.
KING: Dickerson saw that happening, but he didn't work in the clothing business. He was a professor and an entrepreneur. He actually started a few companies. One did computer vision and wound up getting bought for a lot of money. He also had ideas that were less successful.
GOLDSTEIN: Did you get a patent on a self-making bed?
DICKERSON: Yes, I did.
DICKERSON: But I've never commercialized that, even though I think it's a great one to do.
KING: While Dickerson was working on his self-making bed and his computer vision and whatever else, other people were trying to automate sewing, especially people who wanted to keep sewing factories in the U.S. and in other developed countries.
DICKERSON: There was all kinds of efforts - I assure you there were all kinds of efforts - to try to automate it.
GOLDSTEIN: But it was a really hard technical problem to solve for a pretty simple reason.
DICKERSON: Fabric, by definition, is very flexible. It stretches fairly easily, but it distorts in shear very, very easily, and it bends very easily.
GOLDSTEIN: Is distorts in shear a fancy way to say wrinkles?
DICKERSON: Sort of (laughter).
GOLDSTEIN: Not really, actually. I think he was just being nice to me there.
KING: He's a nice guy.
GOLDSTEIN: Like, distorts in shear, it means, like, shifts off to the side kind of. Although it does wrinkle. In my defense, it does wrinkle, too.
KING: And that is a problem. Robots are incredibly good at dealing with things that are rigid, right? You can give a robot a giant piece of steel and tell it to cut some incredibly complicated, precise pattern. So go one millimeter up and two millimeters over, now cut for four millimeters, now go one and a half millimeters to the right, and the robot will get it right every single time.
GOLDSTEIN: Now try and do that exact same thing with a piece of fabric. You know, go two millimeters in from the edge and sew a seam that's 12 millimeters long. But then the fabric starts going into the machine. It does the thing that fabric does. It wrinkles or it folds or something, and the machine doesn't know that. So what you end up with is this seam that's too long or too short or there's some messed up wrinkle that gets sewn into the fabric, which is why you need a human being sitting there at the sewing machine, looking at the fabric and guiding it through.
KING: How do you solve that problem? Machine learning? 3-D printing?
DICKERSON: The most common method that seemed to come close to working was to take the fabric and, let's say, starch it very severely so that, in fact, it became like a piece of sheet metal.
GOLDSTEIN: If, like, robots are good with metal, so let's turn this cotton into something that's like metal.
DICKERSON: That's exactly right.
KING: OK, that solves the bunching and wrinkling problem. But there are a lot of steps in sewing clothes where you want to be able to bend or fold the fabric a little bit, say, when you're sewing a sleeve onto a T-shirt. That is three-dimensional. And if everything is starched really flat, it's just not going to work. So starching was not a solution to everything.
GOLDSTEIN: Some parts of the process of making clothes did get automated over the past few decades. Like cutting fabric into pieces, that is apparently pretty automated now. But, you know, over the same time period, transportation was getting cheaper, trade barriers kept getting lower, and technology kept making it easier to outsource sewing to low-tech factories in Asia and Latin America. So Dickerson says people decided it just was not worth it in the end to automate most kinds of sewing.
DICKERSON: So by the year 2000, people had given up.
KING: A few years ago, Dickerson was semi-retired and he was asked to give a speech on the future of robotics.
DICKERSON: They asked me to do the talk and said - in my own mind, I said, well, what would be a good problem to solve by robotics that hasn't been solved already? And one of them, of course, from my background and the family's background, was sewing.
DICKERSON: And it's one of those things - I've had it happen many times to be truthful with you - all of a sudden, you get an inspiration. And it's the kind of thing that happens in just a few minutes if you're lucky. I don't know whether it was the first few minutes or the - an hour into it or the third or fourth time I came back to the problem.
GOLDSTEIN: But whenever it was, here's the big idea that popped into Dickerson's mind, something better than starch - instead of trying to change the fabric, change the robot. Dickerson realized you could essentially make a map of every single thread in a piece of fabric and then put cameras on the robot so that the machine could track where those threads are at every moment.
DICKERSON: It is counting the threads at the needle...
DICKERSON: ...As they pass. You have to take a thousand pictures per second. And you process that image a thousand times a second, and you get a count of the threads that have passed.
GOLDSTEIN: So now constantly, every thousandth of a second, the robot is like, where's the fabric now? Where's the fabric now, you know? So if it starts to stretch or bunch up or shift, the camera will instantly see that, and it'll send a signal to, you know, some robot hand or whatever and say, hey, fix the fabric.
KING: After he gives that talk describing his idea, Dickerson gets a patent on it. And he thinks, if I can actually make this work, I could bring industrial sewing back to the U.S., not just for military uniforms but for civilian clothes like T-shirts and jeans.
GOLDSTEIN: He joins up with a few other people, and they decide to start a company.
DICKERSON: We got together and we just sat around a conference room and said, what should we call this company? And that's where the name SoftWear came out.
KING: SoftWear - W-E-A-R - like clothes. It's a pun.
DICKERSON: Yes, and software is how we run it.
GOLDSTEIN: I - no, I get it.
DICKERSON: It runs on software.
GOLDSTEIN: I get the joke. I get it.
DICKERSON: (Laughter) OK.
KING: They got the DARPA grant in 2011, got some venture capital in 2014. SoftWear Automation is now a real company.
GOLDSTEIN: And I went down to visit. It's down in Atlanta in this little one-story brick building. It's a few buildings down from an auto shop, across from, like, some kind of vacant lot, next this little weathered church house. Steve Dickerson is not involved so much in the day-to-day operations, so a guy named Barry Clark showed me around. He's the company's head of research and development.
What are these things? There's - I don't even know what to say. What are these things here?
BARRY CLARK: So they're - we call them sewbots (ph). So they're, you know, robots that sew.
GOLDSTEIN: So these are the robots.
CLARK: These are the robots. That's right.
GOLDSTEIN: We're in the company's demo room. There's a concrete floor, brick walls, and we're standing in front of this big table that has two separate machines on it. One machine is this robot arm, and on the end of the robot arm, there's this flat plastic rectangle - think of, like, a little bit bigger than a cafeteria tray, I think - and it's got holes on it.
CLARK: It has a vacuum attached to it which actually allows us to pick up the fabric and move...
GOLDSTEIN: OK, a vacuum. I know what that is.
CLARK: There you go. That's right. So it can suck. And so what we do is we use it to actually pick up the pant leg and place it, and then place the other pant leg on top of it.
GOLDSTEIN: Oh, I got it. So it's like when you have just the tube thing on your vacuum cleaner and you can use it to, like, pick something up. It's like that.
CLARK: That's exactly right, yeah.
GOLDSTEIN: So that's one machine on one side of this table. On the other side, there's another machine, and that one actually looks a lot more familiar.
CLARK: This began its life as an industrial sewing machine.
GOLDSTEIN: And it's got the - it's got the needle, it's got the thread up here. It looks like a sewing machine.
CLARK: That's right. It's not too alien.
GOLDSTEIN: Clark and his team have added these little metal plates right next to the needle. They basically replace the hands of the seamstress. They guide the fabric through the machine.
So you got the vacuum thing on a big crazy arm and then you got the sewing machine part.
CLARK: That's right.
GOLDSTEIN: Though it seems like those are the two main parts.
CLARK: So really the most important part is the computer vision aspect of it, which is the camera which replaces the eyes.
GOLDSTEIN: That is actually Steve Dickerson's big breakthrough, the camera - or cameras, there's actually three of them - that count every thread in the fabric and keep track of it. So those are the different parts of the machine. Now, Clark says, he's going to show me how the whole thing works. The robot is going to sew together two pieces of denim to make a leg of a pair of jeans. And then, it happened, the whole deal, the robot arm, the vacuum...
(SOUNDBITE OF ROBOT VACUUMING)
GOLDSTEIN: ...The sewing.
(SOUNDBITE OF ROBOT SEWING)
GOLDSTEIN: And this part, the sewing, it looks just like a pair of jeans going through a sewing machine except there's nobody sitting there, nobody to push the fabric through.
CLARK: And now we finished sewing the piece, and so we eject it.
GOLDSTEIN: Clark says if this were a factory instead of a demo room, there would be lots of these machines doing each step. So, you know, it would be like vacuum, sew, vacuum, sew, until you had an entire pair of jeans sewn by robots.
KING: OK, so, to be clear, the robots can't do this yet. There are still some problems that Clark and his team are working on, like zippers, for example.
GOLDSTEIN: It's a big problem.
KING: Big problem. But the company has been selling robots that can sew simple things.
GOLDSTEIN: In fact, just the day before I visited, they made a deal to sell robots to a company making towels in Thailand.
KING: And they're close to selling robots that can make T-shirts.
GOLDSTEIN: I talked to one potential customer, Dov Charney.
Do you think you're going to buy machines from SoftWear Automation for your new company?
DOV CHARNEY: Absolutely.
GOLDSTEIN: Dov Charney, as you may know already, founded American Apparel. There was all this controversy there. He left. Now he is starting a new company. He plans to manufacture shirts in Los Angeles using machines from SoftWear Automation.
CHARNEY: I think they have the magic to make things happen.
KING: But, he says, it's not like you can flip a switch and there will be a robot army making T-shirts tomorrow.
CHARNEY: This is probably year one on a 20-year journey.
GOLDSTEIN: So what does it look like in year one?
CHARNEY: What it looks like year one is the machine's not working right and we're pulling our hair out.
GOLDSTEIN: Still, if SoftWear does put all the pieces of its robot army together, it's pretty easy to imagine factories that sew clothes sprouting up all over the U.S. Again, here's Steve Dickerson.
DICKERSON: That's almost a guarantee. If we can meet our targets on cost and performance, it will happen.
GOLDSTEIN: And so those factories will not look at all like the factories that closed up 40 years ago.
DICKERSON: That is correct. However, if you took the broad picture and looked at the factory production floor, the only thing that would be different is very few people, if any.
GOLDSTEIN: Yes, that is a - not a trivial difference. There used to be a person running the sewing machine, and now it will be a computer.
DICKERSON: That is correct.
KING: You will need some people to work in the factory, mostly to take care of the robots. So this is fewer jobs requiring a fair bit of education and probably paying pretty decent salaries. This is what manufacturing looks like in America now. It's not factories where you can walk in the door straight out of high school with no real training and have a solid, lifelong job. Jobs like that are not coming back.
GOLDSTEIN: So that is the U.S. side of the story. That is manufacturing in America today and in the future. But there's another big question here, and that is what do sewing robots mean for all of those millions of people in Asia and Latin America who sew jeans and T-shirts for a living? Those jobs look very unappealing from here. You know, they're low-paid, the work is repetitive. But they are a huge deal for people in countries that are just starting to industrialize. Those jobs are a first step out of rural poverty, and they have been for centuries. Those jobs might go away too.
KING: Maybe there will be new jobs that we can't even imagine yet. That's what always happened before.
GOLDSTEIN: Yeah, I mean, I think the classic example of that is farming, is agriculture. You know, the majority of jobs in the United States used to be jobs in farming. Almost all of those jobs were destroyed by tractors and other new technologies that made farming way more efficient. Today, like, 1 percent of Americans are farmers. That did not lead to mass unemployment, you know? It meant that food got cheaper. It meant people had more money to spend on new kinds of stuff and on services that didn't exist before. And that created new jobs.
KING: So that might happen or it might not.
GOLDSTEIN: Maybe not this time.
KING: Barry Clark, the R&D guy at SoftWear, says maybe computers and robots will get so good that they'll be able to do all the jobs.
CLARK: If we continue to progress in the way that we're going, I think that's a legitimate possibility that we - that we're completely replaced.
GOLDSTEIN: Noel, as you listen to him say that just now, you actually gasped and put your hand on your chest.
KING: Yeah. This sounds really dark.
GOLDSTEIN: Yeah. I mean, it's very striking. When you talk to people like Barry Clark, people who are working on robots, they are way out there. They are thinking about, you know, what this robot future is going to look like.
KING: It's a world where robots and machines will be giving us all of this material wealth, all of this stuff. They're going to be able to give us everything we want. We just won't have that much to do.
(SOUNDBITE OF MARC FERRARI, RYAN CURRY FRANKS AND SCOTT NICKOLEY SONG, "NOTHING TO SAY")
KING: Send us an email. We're email@example.com. Or you can find us on Facebook or Twitter.
GOLDSTEIN: If you're looking for something else to listen to, try Fresh Air but as a podcast. Same show, same Terry Gross, but you can listen to it whenever you want. You can find the Fresh Air podcast at npr.org/podcast, on the NPR One app or anywhere you get your podcasts.
KING: Our show today was produced by Sally Helm.
GOLDSTEIN: It was also edited by our new editor, Bryant Urstadt. Welcome, Bryant.
KING: Welcome. I'm Noel King.
GOLDSTEIN: And I'm Jacob Goldstein. Thanks for listening.
NPR transcripts are created on a rush deadline by Verb8tm, Inc., an NPR contractor, and produced using a proprietary transcription process developed with NPR. This text may not be in its final form and may be updated or revised in the future. Accuracy and availability may vary. The authoritative record of NPR’s programming is the audio record.