Making Robots That Can See

Sep 17, 2012Last month, I read an interesting article published by The New York Times regarding a new generation of robots that employ vision technologies to perform manufacturing tasks more quickly and more accurately than humans possibly could (see Skilled Work, Without the Worker). The writer, John Markoff, describes a Philips Electronics factory on the coast of China in which hundreds of workers use their hands and specialized tools to assemble electric shavers, and another factory in the Netherlands in which 128 robot arms perform the same work with yoga-like flexibility. "Video cameras guide them through feats well beyond the capability of the most dexterous human," the article explains.

Markoff goes on to say that "rapid improvement in vision and touch technologies is putting a wide array of manual jobs within the abilities of robots," and he discusses the implications for employment in manufacturing during the coming years. I have long believed that robots would increasingly be able to do many of the jobs humans do today, but I thought it would be radio frequency identification technology that would transform these machines.

One challenge facing robots has been flexibility. They could be built to perform a single task repeatedly, but not to distinguish between, say, a red or blue bicycle seat to be put on a frame. A tag with a unique serial number could be attached to each seat, while a reader on the robot's arm would allow it to easily distinguish between the two. The Times article suggests that vision technologies might make that possible as well.

Does that mean RFID will not be used? I've been pondering that question for the past few weeks, and here's what I've concluded. Robots will, through the power of computer processing, increasingly attain the capabilities of humans. So if a person can execute a task effectively using their eyes, robots will likely utilize vision technologies. If humans cannot use their eyes, then robots won't be able to use vision.

Let me provide a couple of examples to explain.

A human worker on a factory line could easily distinguish between a red and blue seat using his or her eyes. So it's likely a robot would utilize vision technology to determine which seat to pick. However, a human can't easily identify a specific car chassis on an assembly line and know which specific parts should be attached to it. At present, a worker would likely scan a bar code or have an RFID tag read automatically. In such cases, robots would likely do the same. Humans can't tell the difference between two pairs of jeans folded on a store shelf, and neither would a robot. A human also can't count 100 items within a box without opening that box, and neither would a robot.

I have long believed that robots and radio frequency identification would open the door to customized manufacturing. Car manufacturers currently limit customers to a couple of styles of each car they offer—blue with a beige interior, for example, or burgundy with a black interior—because customization increases costs. If items can be tracked via RFID and be assembled by robots, a customer could theoretically have a company design a car on the line and have it built to order at little extra cost. I still believe that is where the world is headed, and that manufacturing will eventually move from low-wage countries back to where the market is, enabling customized goods to be delivered more quickly.

It will be fascinating to watch how this plays out—and, perhaps, a little scary for those currently working at low-wage assembly jobs.

In the meantime, you can read about the use of RFID with robotics in two recent Expert Views: Humanizing Robots and Enlisting Robots.

Mark Roberti is the founder and editor of RFID Journal. If you would like to comment on this article, click on the link below. To read more of Mark's opinions, visit the RFID Journal Blog, the Editor's Note archive or RFID Connect.