This segment is part of the special report "Engineers of the New Millennium: Robots for Real."
Transcript: Lessons From a Mechanical Child
Reported by Giselle Weiss
Giselle Weiss: A baby is a great explorer. Infants gnaw, lick, grab, prod, and crawl their way into awareness of the world and of themselves. But how do they do it? That's the question posed by RobotCub, an initiative whose goal is to study cognitive function by building a humanoid, childlike robot, known as iCub. The project is funded by the European Union and involves 10 academic partners. So far, eight robots have been built. Here at the Swiss Federal Institute of Technology in Lausanne—EPFL for short—researchers are working on systems that control iCub's upper and lower body.
Giselle Weiss: Auke Ijspeert heads the Biologically Inspired Robotics Group at the EPFL. He and his coworkers specialize in locomotion. They're programming the robot to get from A to B not on two limbs but on four—by crawling, as an infant does. It's no easy feat.
Auke Ijspeert: If you have a wheeled robot, it's very simple to control its locomotion. But if you have a humanoid robot, you have to move maybe, say, seven joints in each leg plus seven joints in the arm, so that means 28 joints that you have to coordinate properly.
Giselle Weiss: Full-fledged iCub robots were developed in 2004. Before that, the scientists used computers to explore principles like gravity and friction that are important in crawling. Then they tested simple arm movements on an existing robot by having it pound a drum.
Giselle Weiss: Eventually, in 2008, a robot with legs became available in Genoa, Italy, and one of Ijspeert's students took it for a crawl.
Giselle Weiss: The robot moved a little clumsily and even bruised its arms. But the scientists are helping iCub to clean up its crawling technique.
Auke Ijspeert: Something we haven't done yet so far is dealing with complex terrain. So, for instance, going up or down a sofa, or going down stairs, climbing up stairs. These are very complex problems, also in terms of visual processing.
Giselle Weiss: Several buildings away, Ijspeert's colleague Aude Billard is also working on iCub. Billard, who leads EPFL's Learning Algorithms and Systems Laboratory, is out of the country, and her student Eric Sauser shows me around. The robot is anchored vertically to a workbench, apparently snoozing.
Giselle Weiss: The researchers are interested in three major questions: How does the brain build a model of its own body? How do you coordinate the motion of your eyes, head, and arms to reach an object? And how do you use that information to learn from imitating somebody else? Sauser straps some sensors to his left arm and hand and powers up iCub. As he moves his arm and fingers, the robot does likewise. Like us, it has an opposable thumb, which enables it to grasp objects and release them.
Eric Sauser: We had a screwdriver before. That may be a bit scary, a robot with a screwdriver. But I'm confident, since it's me that controls the robot. So here the robot got the screwdriver. He can move it around in the air, hit the floor.
Giselle Weiss: The scientists will also study the kind of nonverbal information we can only get through touching, as when a tennis coach lifts the elbow of a player to correct a stroke. Eventually, Billard and Ijspeert hope to combine their efforts to study more complex gestures. But, as Ijspeert explains, that's still a while off.
Auke Ijspeert: We are now just at the beginning, in terms of the software. So now we have a nice robot. But there's much more work to do in terms of really addressing questions of what it means to have some artificial intelligence.
Giselle Weiss: Perhaps someday, like any kid, iCub will be able to clean its room or to play Ping-Pong with you. For now, it's not half bad on the foot-pedal drums.
Giselle Weiss: In Lausanne, Switzerland, I'm Giselle Weiss.