Robots and infants, and spoons, oh my
| HOOPING IT UP: Justus Piater, James Davis, and Andrew Fagg step out with Magilla, the Hula-Hooping robot. (Ben Barnhart photo) |
The first time I saw Magilla, he was twirling a Hula-Hoop on his arm to an enthusiastic crowd. The occasion was the September 14 open house at the new UMass Engineering and Computer Science Research Development Center. Over 400 people attended and viewed the exhibits and demonstrations in the three-story, $14.6-million building. In between Hula-Hooping, I could have sworn Magilla did a robot version of the Macarena.
Robot version?
|
The new computer science building (Stephen Long photo) |
Magilla is a six-foot-tall robot that research scientist Andrew Fagg and graduate students James Davis, Justus Piater, and David Wheeler are using as part of their work on artificial intelligence in the computer science laboratory for perceptual robotics, directed by Professor Roderic Grupen. In collaboration with psychology professor Rachel Clifton and research associate Michael McCarty, their aim is to build computer models based on the learning process of infants. In one experiment, they present a nine-month-old infant with a spoonful of applesauce, either bowl or handle first. “If it’s the wrong configuration, the infant will get the handle in her mouth rather than the applesauce,” says Fagg. “At that point, she’ll realize something is wrong and go through contortions to fix it.” Over the next few months, he says, the infant begins to anticipate. At fourteen months she’ll still grab it wrong, but realize it and take corrective action before it gets to her mouth. The researchers think that understanding this learning process will be useful in teaching robots to perform tasks.
They plan a robot experiment that parallels the applesauce one, although its exact form is undecided. They want to help the robot learn to choose actions such as grasping, and to anticipate how these will affect subsequent actions, such as placing an object inside something else.
Will teaching Magilla to twirl a Hula-Hoop eventually lead to even higher forms of artificial intelligence? Grad student James Davis explains the connection this way: “The robot adjusts the frequency of the arm movement just as humans do, so that it feels good, so it has the timing right and gets the right push.” We humans don’t even think about the fact that we can spin a yo-yo, turn a crank, work with a Slinky. We don’t have to concentrate on the fact that there’s an “oscillatory behavior” and we’re adjusting how it feels on our hands or on our arms. It’s just learned through development, says Davis.
Now, I wondered, can we get Magilla to eat applesauce?
— Linda Cahillane
