The robot watched as researcher Shikhar Bahl opened the refrigerator door. He fixed it
It didn't work at first:sometimes he missed the handle entirely, grabbed it in the wrong place, or pulled the wrong way. But after several hours of practice, the robot succeeded and opened the door.
"Imitation– great way to learn", – said Bahl, a graduate student at the Robotics Institute (RI) at Carnegie Mellon University's School of Computer Science.
Bahl worked with Deepak Pathak and AbhinavGupta, RI faculty, to develop a new learning method for robots called WHIRL. This is an efficient algorithm for instantaneous visual simulation. It can learn directly from video footage of human interaction and generalize that information to perform new tasks, making robots well-suited for homework training.
People are constantly performing various tasks in theirhouses. With WHIRL, the robot can observe these tasks and collect the video data it needs to ultimately determine how to complete the job itself.
The robot has learned to perform more than 20 tasks –from opening and closing cabinet doors and drawers to putting a lid on a pot, pushing a chair, and even taking a trash bag out of the trash can. Each time, the robot watched a person perform a task, and then trained and learned to perform it on its own.