Robots at Work: Toward a Smarter Factory
By Rodney Brooks
Many fear that a robotic takeover of manufacturing jobs will keep humans out of work. But one inventor shows how tomorrow’s manufacturing robots will be smaller, smarter, and co-worker friendly—and they’ll let manufacturers stop chasing around the world for low-wage workers.
In the next forty years, the world will be transformed by demographics. By 2050, the ratio of working-age people to retirees in Europe, much of Asia, and the United States will be very different from what it is today. The United States is also about to undergo a massive increase in the population older than 65, so policy makers are worried about how that will affect Social Security.
But this shrinking of the working-age population doesn’t just affect national budgets. It affects services. There will be fewer workers to provide services to older people. More competition for those services means prices will go up. So productivity per worker will have to go up. To do that, we will need more robots to help people be more productive. This is why industrial robotics has become my focus.
I became concerned about the lack of innovation in industrial robotics after spending time in Shenzhen, China, as we set up the production line for the Roomba vacuum cleaner. I saw people building a million robots a year and doing it by hand. This isn’t unusual in electronics manufacturing. Consider that the iPad is touched by 325 pairs of hands during assembly. That means that, despite growing interest and anxiety about automation and robotics taking away manufacturing jobs, most of our stuff—the low-cost consumer items that we buy from Wal-Mart—is still made by hand.
People sometimes think manufacturing is dead in the United States. In fact, manufacturing activity comprises a $2 trillion portion of the U.S. economy, very similar to the dollar value it comprises in China and Europe. In Japan, that number is about $1 trillion.
The United States has kept that manufacturing activity by increasing worker productivity, which has gone up about 3.7% per year for 60 years—a good run. The United States has kept the higher-value-added manufacturing and let the lower-value-added manufacturing go elsewhere. And the definition of “elsewhere” has changed over time. The manufacture of simple goods is constantly moving to the location with the lowest wages.
Ethically and environmentally, this is a complex and controversial subject. I argue that—from the perspectiveof business—it’s unsustainable.
After the end of World War II, there was an abundance of low-cost labor in Japan, so that’s where manufacturing moved. But as the Japanese economy recovered and the standard of living went up, the cost of making goods also went up. So low-cost manufacturing moved to South Korea. After “the miracle of South Korea,” it moved to Taiwan. When the standard of living went up there, manufacturing moved to the province of Shenzhen in China.
I used to be part of a company that manufactured toys in China, but by the late 1990s, much of the sewing work my company required for robotic dolls we were selling was being done in Vietnam. We now see all sorts of other low-wage and low-skilled jobs moving to Vietnam.
Businesses are constantly chasing this low-cost labor—people who are willing to do difficult work producing low-value products. As the standard of living goes up in these places, education expands. That’s a good thing. But from the perspective of a manufacturer, a more intelligent and skilled workforce simply has less interest in these types of jobs. I’ve seen it myself. In Shenzhen, where my previous company, iRobot, was building Roomba robot vacuum cleaners, we tried to add another line. The factory manager said, “All my workers went off to college.”
Where will manufacturing activity go next? Thailand has had a tremendous industrial revolution in the last 30 years, as have Indonesia and Malaysia. Eventually, we will run out of places where there is low-cost labor.
For U.S. companies, making products closer to home, where they were designed, allows you to keep your supply chain short and responsive. You can produce and ship closer to where your engineers are located, so you can innovate designs or develop new ones faster. You can tighten the loop between innovation and manufacturing. Your design-to-production cycle quickens, and you have fewer intellectual and property concerns. These are just some of the reasons for a company to keep its manufacturing operation close to where the rest of its operations are located.
The question is: What will it take for us to break out of the cycle of making cheap stuff by hand, where companies make expensive, value-added merchandise with machines and automation, but low-value products like toys with unskilled, inexpensive labor? This is a challenge my company is trying to solve with robotics.
A New Robot for a New Robot Economy
The first industrial robot developed in the United States went to work in 1961 in a Ewing, New Jersey, GM factory. Called the Unimate, it operated with a die-casting mold placing hot, forged car parts into a liquid bath to cool them. At the time, you couldn’t have a computer on an industrial robot. Computers cost millions of dollars and filled a room. Sensors were also extremely expensive. Robots were effectively blind, very dumb, and did repeated actions following a trajectory.
If you’ve watched the evolution of industrial robots over the last 50 years, you haven’t actually seen much innovation since the Unimate. These machines perform well on very narrowly defined, repeatable tasks. But they aren’t adaptable, flexible, or easy to use. Nor are most of these machines safe for people to be around.
Today, 70% of the industrial robots in existence are in automobile factories. They’re either in the paint shop or the body shop. Go to a car factory in Japan or Detroit, you’ll see a body shop full of robots, but with no people. Go to final assembly and you’ll find all people, no robots. Industrial robots and people don’t mix.
These machines are often heralded as money savers for factory owners and operators. But the cost to integrate one of today’s industrial robots into a factory operation is often three or five times the cost of the robot itself. It’s a job that demands programmers, specialists, all sorts of people. And they have to put safety cages around the robots so that the robots don’t strike people while operating.
Unlike human workers, who can detect when they’re about to hit something with their eyes, ears, or skin, most of these machines have no sensors or means to detect what is happening in their environment. They’re not aware. All of this speaks to a larger and fundamental flaw with the way factory bots are built today.
In an increasingly interconnected world, industrial robots have not followed the information technology revolution. In our march to the future, we somehow left robots behind. Some colleagues and I decided to change that.
We saw 300,000 small manufacturing companies in the United States with fewer than 500 employees. Almost none of these firms have an industrial robot, for some of the reasons outlined above. Almost all of these firms have relatively small production runs. That means they’re constantly changing the design and manufacturing procedures for what they produce.
Some of these companies produce a wide variety of goods for other companies. They’re what are sometimes called job shops. They specialize in manufacturing a type of product that can be highly customized to an individual client’s needs. In a typical factory with an industrial robot, a production run is rarely less than four months. For these job shops, a run can be as short as an hour. In these 300,000 very small companies, we saw a large potential market.
In September 2008, we started our company, Rethink Robotics, backed with capital from Charles River Ventures, Amazon’s Jeff Bezos, and other Boston and Silicon Valley players. We announced our first product launch on September 18, 2012. And we’re just starting to ship now. We build a robot named Baxter, a new type of industrial robot that sells for $22,000.
Baxter is very different from existing industrial robots. It doesn’t need an expensive or elaborate safety cage, and factory operators don’t need to put it in a part of the factory where it’s segregated from the rest of the workers. It’s safe to share a workspace with. If you see a Baxter, you can actually go and hug it while it’s in operation.
Baxter also works right out of the box. Typically, it takes 18 months to integrate an industrial robot into a factory operation. With Baxter, it’s about an hour. Baxter requires no specialized programming. A factory floor worker with only a high-school diploma, someone who has never seen a robot before, can learn to train Baxter to do simple tasks in five minutes. If you’re a manufacturing engineer, you can go deeper into the menu system and adjust and optimize settings for different tasks. More-complex tasks require a bit more training.
Interacting with Baxter is more like working with a person than operating a traditional industrial robot. If the robot picks up something it shouldn’t on the assembly line, for instance, you can take its arm and move the robot to put the object down. (Don’t try that with a current industrial robot.) You don’t have to know anything about computer languages. It knows what you mean, and it does what you want.
Baxter is outfitted with a variety of sensors, including depth sensors as well cameras in its wrists, so it sees with its hands. It’s constantly building and adjusting a mathematical model of the world in front of it, allowing it to recognize different objects.
We made the robot simple and easy to use because we don’t think that it’s the customer’s job to do costly engineering work on a product that he or she has purchased. We wanted to build a machine that was as intuitive to use as the iPhone.
This means that ordinary factory workers are now empowered to be the programmers. They’re not in competition with these machines, because an average factory worker can serve as a supervisor to Baxter. A factory worker can show the robot a fragment of the task she is asking the robot to perform, and the robot infers the rest of the task. And if someone is interacting with the robot or doing part of the task, Baxter can figure out to just do the rest of the task.
We built Baxter for manufacturing, because in manufacturing we saw a good return on investment. But Baxter’s potential extends beyond factory work: This is going to revolutionize the use of robots in research settings, as well. In the 1990s, the availability of low-cost mobile robots allowed thousands of researchers to work on a problem known as SLAM, Simultaneous Localization and Mapping. That led ultimately to the Google self-driving car. Now for the first time, thousands of researchers will be able to afford two-armed robots, which we think will lead to similarly important new research. Some of that research will seem crazy, silly, or even dumb, but out of that we will see new applications for two-armed robots interacting with people.
These applications exist in health care and eldercare, and a wide variety of other areas. We’ll find thousands of uses for this robot. We’ve only just begun.
About the Author
Rodney Brooks is a computer scientist and co-founder, in 1990, of iRobot, makers of the Roomba robot vacuum cleaner. He is the founder and chief technical officer for Rethink Robotics, www.RethinkRobotics.com.
This article was adapted with the author’s permission from a presentation at the 2012 EmTech conference, which took place at the MIT Media Lab October 24–26, 2012. Learn more at www2.technologyreview.com/emtech/12/.
KEEP UP WITH WFS NEWS & UPCOMING EVENTS
August 2, 2015 - Tomorrow the United States will announce regulations to cut greenhouse gas (GHG) emissions from power utilities nationwide.
August 1, 2015 - The Monsanto patent on Roundup Ready Soybeans has expired after 20 years. That means farmers can now legally collect the seeds from their plantings and use them to replant or even sell to other farmers.
U.S. Study Shows Oil Sands Produce 20% More Carbon Than Conventional Crude Heating Up Keystone XL Debate
July 31, 2015 - Today United States crude production includes 9% contribution from Canadian oil sands. By 2020 that number is expected to climb to 14%. This estimate is independent of whether the northern leg of the Keystone XL pipeline gets built or not. So the United States has much invested in Canadian oil sands production and as a result the U.S.
July 30, 2015 - One of my readers who follows my blog through LinkedIn admonished me a couple of weeks ago for being too critical of the fossil fuel industry.
July 28, 2015 - If you are a regular reader of this blog then you have read about the importance of our research into stem cells and their therapeutic value. I'm even contemplating having my stem cells harvested to inject into my osteoarthritic left knee to help restore the cartilage I have lost over the years.
July 26, 2015 - Three farm stories caught my eye this week. The first, a truly revolutionary one that pushes back the dawn of the age of agriculture some 11,000 years. The second, a GMO story featuring a new rice that produces less greenhouse gas. And the third, a Harvard study about declining zinc levels in food because of rising carbon dioxide levels in the atmosphere.