'Robo Brain' Helps Teach Robots All They Need to Know from the Internet

Robo brain--otherwise known as a large computational system that learns from publicly available internet resources--will help robots draw on new knowledge and sharpen previously existing skills.

Researchers at Cornell University, Stanford and Brown University of California, Berkeley have worked to teach the robots certain skills one at a time, such as finding an owner's keys, pouring a drink or even putting the dishes away.

"Our laptops and cell phones have access to all the information we want. If a robot encounters a situation it hasn't seen before it can query Robo Brain in the cloud," said Ashutosh Saxena, assistant professor of computer science at Cornell University, in a news release.

By connecting images for Robo Brain with objects, video and text to help robots properly learn a task or action, the researchers have helped them recognize and identify certain objects, along with general human language and behavior.

For instance, if Robo Brain sees a coffee mug, it can learn not only that this object holds coffee but that liquids can be poured into it or out of it by simply grasping the handle.

The Robo Brain system incorporates "structured deep learning," where information can be stored in many levels of abstraction.
A robot's computer brain will then store what it has learned in the form that mathematicians refer to as Markov model, which graphically sets points connected by lines that project the image or action.

"The Robo Brain will look like a gigantic, branching graph with abilities for multi-dimensional queries," said Aditya Jami, a visiting researcher at Cornell, who designed the large-scale database for the brain. 

More information regarding the findings were presented at the Robotics: Science and System Conference in Berkeley, Calif.