Inexpensive personal robots could be available soon, but they first have to be able to learn to perform tasks that human know how to do instinctively, says Iowa State University in Ames professor Alexander Stoytchev. A truly useful personal robot needs to be able to learn on its own through physical and social interactions with its environment, says Stoytchev, who is developing software that enables robots to learn. His research team has already succeeded in giving a robot the learning abilities of a two-year-old child.
In one set of experiments, the robot was presented with 36 different objects and asked to identify and classify the objects based on the sounds they made, with the robot grasping, pushing, tapping, shaking, and dropping the objects. After a single action, the robot had a 72 percent success rate. The robot's accuracy increased after each successive action, reaching 99.2 percent accuracy after all five actions.
The robot had to learn to use a perceptual model to recognize and classify objects, which it could use to estimate how similar two objects were based on the sounds they made.
When robots are ready to serve as personal assistants, they may resemble the Home Exploring Robotic Butler (HERB) prototype developed at an Intel lab as part of the company's Personal Robotics Project. HERB is a three-foot machine that balances an arm and a hand on top of a Segway personal transporter base. HERB uses laser range finders and a camera to navigate its environment, and uses learning algorithms and probability distributions to learn how people move and avoid running into them.
From Scientific American
View Full Article