Disney Research Pittsburgh scientists are aiming for human-like robots
May 28, 2013 4:00 AM
Disney Research associate Matthew Glisson hands a bag to a robot during a simulated demonstration at the Disney Research lab on the Carnegie Mellon University campus in Oakland. Researchers have developed a robot that is capable of analyzing human motions in order to give and receive items.
By Jessica Tully Pittsburgh Post-Gazette
Many viewers of the 1962 animated television series "The Jetsons" scoffed at the idea of robots folding clothes or tending to the kitchen, imagining that this type of futuristic technology would not be available for hundreds of years.
But handing coats, shoes or packages to a robot could soon become customary, if Disney Research Pittsburgh's latest development is introduced to the workplace or the home.
Scientists at the Pittsburgh lab have created a humanoid robot that can receive an object a person hands to it. But the robot doesn't simply stick its hand out blindly -- that type of movement is more robotic, less human, and thus, less friendly, researchers decided.
"I was interested in making the robot more physically interactive instead of working on emotional interaction," said Katsu Yamane, a senior research scientist for Disney Research here. "I wanted to see the robot act human-like."
This project was partially funded by the International Center for Advanced Communication Technologies at Carnegie Mellon University and Karlsruhe Institute of Technology in Germany. Findings were presented at an international conference in Karlsruhe, Germany, in early May, and the project essay was nominated for the Best Cognitive Robotics Paper Award, Mr. Yamane said.
In creating a more human-like robot, two challenges needed to be addressed: teaching the robot how to recognize when a person hands it an object and teaching it how to predict where the person intends to make the handoff.
Mr. Yamane was joined by Marcel Revfi, an exchange student from Karlsruhe Institute of Technology, to find the answers.
The pair used motion-capture data of two people interacting to create a database of human motions. By quickly searching the database, the robot can judge what the human is doing and where he will likely extend his hand, Mr. Yamane said.
Siddhartha Srinivasa, a professor at CMU's Robotics Institute, said he has seen data recognition used before in computer graphics and animation. What is novel about this development, he said, is that it was applied to human-robot interaction.
He called it the first step in allowing robots and humans to collaborate seamlessly and fluidly.
"Those of use who work with robots have to think more and more about how a robot's motion can be more functional and expressive," Mr. Srinivasa said, adding that he would like to soon teach a robot how to prepare a meal or clean off the dinner table with a human.
While scientists focused primarily on studying handoff motions, this database technique developed by Disney Research Pittsburgh also could help robots learn other common human motions, such as dance. Mr. Yamane said he would like to add finger motions and secondary behaviors to make the robot even more engaging.