MIT Researchers Use Old Math Model to Give Robots a Better Vision

Researchers from the Massachusetts Institute of Technology (MIT) have created a new algorithm based on a rarely used statistical model, which can advance robot vision.

Vision isn't limited to just seeing objects; one even needs to recognize them and understand their orientation in space. For example, eyes send visual data to the brain and the brain decodes that information like whether the handle is facing right or left.

Now, MIT researchers have shown that a neglected statistical construct called the Bingham distribution can help robots understand object orientation.

Jared Glover, a graduate student in MIT's Department of Electrical Engineering and Sanja Popovic, an MIT alumna, have shown that the new robot-vision algorithm based on Bingham distribution is 15 percent better than any other program in improving robots' understanding of objects.

The researchers will be presenting their study paper at the International Conference on Intelligent Robots and Systems in November.

Robots Playing Pingpong

Glover and team are using the Bingham distribution to understand the orientation of pingpong balls in space and are applying the information to teach robots to play pingpong (check embedded video). According to the researchers, the model's statistical probability gives a good construct even when dealing with limited data.

"Alignment is key to many problems in robotics, from object-detection and tracking to mapping," Glover said in a news release. "And ambiguity is really the central challenge to getting good alignments in highly cluttered scenes, like inside a refrigerator or in a drawer. That's why the Bingham distribution seems to be a useful tool, because it allows the algorithm to get more information out of each ambiguous, local feature."