Simon the robot has just learned a new skill: transferring a red block from one hand into a coffee cup held by the other. But like an eager preschooler, he wants to know more. “Can I begin here?” he asks, lifting the block high. Simon has two arms, eight fingers, doe eyes, and a monotone voice. With each question and answer, he is doing what roboticist Andrea Thomaz calls “whittling away the hypothesis space,” or eliminating information that is not essential. “The idea is that you tell him something, he asks a few questions, you provide a couple examples, then he builds a model,” Thomaz says. If the exchange progresses smoothly, Simon is soon folding your laundry.
Using Simon as her student, Thomaz is redefining how robots and humans interact. She sees a future where any “naive user” (or nonprogrammer) could buy a robot, take it home, and Andrea Thomaz
Georgia Techinstruct it to do almost anything. But for this to work, robots need to think like naive users. So Thomaz invites folks in off the street to teach Simon at her Georgia Tech lab. Lessons include everything from clearing the dinner table to sorting objects by color. Based on the results, Thomaz tweaks Simon’s algorithms to make him a more efficient communicator and learner.
With a growing number of robots used in health care and with the promise of them one day assisting the elderly and disabled, the implications of robots that can be trained by anyone, in any setting, are huge. “Everyone is different, every house is different, every office, every hospital,” says Thomaz. “I want people to be able to say ‘Hey, robot, here’s how things happen here.'”
Click here to see more from our 11th annual celebration of young researchers whose innovations will change the world