Printable sticker labels have a lot of convenient uses, whether you’re using them to address cheap mailers or to help organize cheap ring binders and other items around your office. However, did you know that high-tech printable sticker labels might someday be used to guide robots? A research team at the Georgia Institute of Technology is trying to turn how we think about artificial vision and more with the help of a fairly commonplace item.
While vision seems like a natural way for robots to identify objects, researchers say it doesn’t work as well in practice as it does for humans. For example, it requires a considerable amount of computing power, and even when that is available, modern robots still aren’t able to use their identification capacity if the object is facing the wrong way or partially-obscured from view.
For this reason, many research teams have considered using RFID tags, defined as wireless transmitters that use radio-frequency identification to enable tracking and identification, to help robots find objects. These tags are extremely cheap, aren’t subject to a number of complications like lighting, and they are printable to boot. However, most attempts have tried to apply the technology in a number of relatively complicated ways, by estimating the printable sticker label’s proximity to the robot or using a data-driven sensor model. Unfortunately, there are too many variables for these methods to be used reliably in the real world.
However, the research team at Georgia Tech had a far simpler idea: using the RFID tags to essentially play the children’s game “Hotter/Colder”.By outfitting a robot with a pair of shoulder-mounted antennas and inputting some simple behaviors, they trained the robot to wander around an assigned search area and make notes of wherever it picks up the radio signals. The robot then goes to the spot where the signal was “warmest”, gauging signal strength with its shoulder antennas to pinpoint the most accurate location. As a result, the system can find tagged objects in a variety of somewhat obscured positions without a complicated visual search.
This approach to identification is less streamlined than the traditional vision idea, but researchers say it is simpler and easier to implement, could actually work outside of a lab, and could be used fairly soon to help people who need extra assistance. Just imagine: you might be putting printable sticker labels on your cheap bubble mailers and numbered dividers today, but in the future, you could be using them to guide a robot.
