Few sectors have adapted to robots as much as manufacturing. But while you can easily find a plant where humans and robots are both working to create a product, they almost never work side by side.
Dr. Julie Shah and others at MIT’s Computer Science and Artificial Intelligence Laboratory think that could change in a matter of years, thanks to a new algorithm.
Most robots in manufacturing are large and can be dangerous to human workers. They’re also fulfilling a static set of commands in an order that’s preordained, not learning from their work. Humans, on the other hand, don’t all approach tasks the same way. Say you have to fill a bunch of holes with glue, then put a nail in those holes, and, finally, wipe off the excess glue. Would you do each one by one—or put glue in every hole, then a nail in every hole, and then wipe each one off? It depends on your disposition. For a robotic helper to assist by, say, handing over tools, adapting to this kind of variation is really problematic, and thus, incredibly expensive.
Professor Shaw has been trying for years to answer the question of how we can make robots that can change their behavior at the whim of their flesh-and-bone colleagues. She’s spent time at Boeing—which is helping to fund her current research—and various other manufacturing companies.
“Typically the way we command just about all robots today is we command them step by step. Sometimes it’s at the lowest level, sometimes it’s at a higher level,” she says, “but it’s primarily explicit commanding. Do we have to do that with robots?”
Maybe not. Working with robots like FRIDA, designed by Swiss robotics company ABB, Shah and her colleagues have been experimenting with robots and humans completing manufacturing tasks together. The human worker did the complicated stuff, while robots do “non-value-added” tasks, like supplying the tools the human needs. More importantly, researchers have used motion-capture suits to teach the robot how different workers go about performing the same tasks. In preliminary lab tests, they managed to reduce a human worker’s idle time by a whopping 80 percent, by having the robot learn the human worker’s workflow and adapt its own workflow to speed things up. In July, they’ll present their findings at an international robotics systems conference in Sydney, Australia.
In the near future, Shah pictures robots and humans completing training sessions together, with the robot learning different humans’ preferred work habits, and then being able to develop an appropriate task plan on the factory floor. She points to her lab tests, in which FRIDA helped human workers complete an aircraft wing-building process called spar assembly. Production efficiency could skyrocket, and the production lines and factory floors themselves could also become more fluid and flexible as a result. Shah thinks that within several years her algorithm could change the way we approach manufacturing, and in less than a decade, factory floors could work completely differently. It might be a long way from the collaborative relationship between Tony Stark and JARVIS, but it’s a start.