Robot Able to Learn as it Goes, Adapting to Human Instructions [VIDEO]

By modifying a robot from Rethink Robotics in Boston, researchers have taught it to work in a simulated supermarket checkout line.

According to those involved, the robot was able to "coactively learn" from its human counterparts, making adjustments even as the action was in progress.

Robotics have been successfully implemented in assembly lines throughout a wide range of industries, effectively and precisely carrying out specific actions over and over. Not every situation is as controlled, however. For example, a personal robot working in someone's home would have to understand the need to handle fruit with more care than, say, canned goods, or to keep a sharp knife away from humans.

Saxena and graduate student Ashesh Jain drew on previous work when they implemented programming that enables the robot to plan its own motions yet still allow a human to intervene, manually guiding its arms in order to fine-tune their movement. They also provided it with an algorithm that enables it to learn a little at a time, slowly refining its movements based on human input.

Over time, the robot has learned to pair a particular trajectory to a specific object, differentiating between cereal boxes and eggs, the latter of which requires more care and should not be lifted far above the counter. Sharp objects, the robot learned, should not be moved in wide swinging movements, but instead held close and away from people.

"We give the robot a lot of flexibility in learning," Ashutosh Saxena, assistant professor of computer science at Cornell University, said in a statement. "The robot can learn from corrective human feedback in order to plan its actions that are suitable to the environment and the objects present."

The research team will present their work at the Neural Information Processing System conference in Lake Tahoe, Calif. later this year.