Growing up in Phoenix, Bryan Russell sometimes daydreamed about getting help with chores: “I thought it would be cool to have a gadget with robot arms that could do the dishes or something,” he says. Of course, there were no such machines, and still aren’t. But unlike then, today Russell’s in a position to help set the stage for this and other useful devices.
A grad student in MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL), Russell works on a capability vital to any truly versatile robot, which is vision. Specifically, he’s trying to craft artificial intelligence systems that meet what is, for them, a tough challenge: to look at things, and categorize them.
“If you ‘show’ a bunch of images to your system,” he says, “it needs to be able to characterize objects in those images — for example, chairs — without being told what a chair is or where to look for it.”
The field is progressing: today’s best AI systems can learn simple object categories like chairs or cars with a high degree of accuracy. But these systems often don’t always work if the objects are lit in an odd manner, say, or are partly obscured. And some living objects, especially those with natural camouflage — leopards, zebras — all-but stymie the systems. Russell’s focus right now is enabling his system to do something we do without thinking about it: recognize the location of one part of an object in relation to others, like a motorcycle’s handlebars and wheels.
“Our current system reflects the assumption that each feature in an object is independent, so it does not take spatial relations into account,” says Russell. “We’re working on that.” The student, a Dartmouth graduate, says attacking that problem will probably be his doctoral project. His plans after he gets his degree, meanwhile, aren’t set but will definitely involve teaching as well as research. “I like connecting with students,” he notes. “When they have that ‘Aha!’ moment, it’s a thrill.”
Russell’s grateful to have launched his MIT career with the aid of a Norman B. Leventhal Presidential Fellowship — a program created in 2001 by Jeffry and Barbara Picower in honor of a leading MIT graduate. The fellowship, he says, gave him flexibility to explore a range of interests in his first year.
He also says that after four years at MIT, he’s gained new insights into the Institute’s strengths. “What impresses me,” he says, “is how passionate people are about their work.”