According to Engadget, Massachusetts Institute of Technology (MIT) has developed a robot that can teach itself to see.
The robot, called “Dense Object Nets” (DON), collects visual data points arranged as coordinates and generates a “visual roadmap.” From there, its system stitches together the coordinates into a single panoramic image. “This enables the system to better and more intuitively understand the object’s shape and how it works in the context of the environment around it,” Engadget reports.
As a result, the DON system enables a robot to “look” at an object, such as a cup of coffee, orient itself to the object, and lift it. DON also enables a robot to spot a specific object out of a pile of other objects.
Engadget says that the DON system uses an RGB-D sensor, which has a combination RGB-depth camera. Then, DON trains itself, in about one hour’s time: “If you want the system to recognize a brown boot, you simply put the robot in a room with a brown boot for a little while. The system will automatically circle the boot, taking reference photos which it uses to generate the coordinate points, then trains itself based on what it’s seen.”
The goal of the DON system is to enable a robot to do “more advanced manipulation tasks,” Lucas Manuelli, the PhD student from MIT who authored a paper on this technology, told Engadget. This means, for example, the technology would allow a robot to see the above cup of coffee, recognize where the bottom of the cup is, and keep the bottom pointed down while lifting it so that the coffee doesn’t spill.
Functionalities like this will come in handy for factory robots and factory workflows, since these types of robots “often need complex part feeders to work reliably,” Manuelli wrote in the paper. However, DON is still in its early stages, and it will be some time before it debuts in factory – and home – settings.