Researchers at Cornell University have created a “Robo Brain” that will teach robots how humans behave naturally. It’s a large-scale computational system that learns from the Internet. It could be used by millions of robots to serve as helpers in our offices, factories and homes. Ashutosh Saxena and Aditya Jami, two India-born scientists, described the project at the “2014 Robotics: Science and Systems Conference” in Berkeley.
Robots can access the Robo Brain through Internet
The Robo Brain will teach machines how to find keys, put away dishes, pour a drink, and when not to interrupt two people having a conversation. Ashutosh Saxena, an assistant professor at Cornell University, said our cell phones and laptops have access to all the information. If a robot faces a situation it has never seen before, it can simply query Robo Brain in the cloud.
Robo Brain is currently downloading and processing over one billion photos, about 100 million how-to documents and appliance manuals, and 120,000 YouTube videos. Researchers said the system translates the information and stores it in a robot-friendly format. It processes photos to pick out objects in them. By connecting videos and images with text, the Robo Brain learns to recognize various objects and how to use them. It is also learning human language and behavior.
Internet users can correct mistakes or improve the Robo Brain
For instance, if a robot encounters a coffee mug, it can tap into Robo Brain to learn that it’s a coffee mug, and that liquid materials can be poured into and out of it. The Robo Brain will also teach robots that a coffee mug is grasped by the handle. And yes, this computational system will have teachers through crowdsourcing.
Scientists employed a system called “structured deep learning” where data is stored in multiple levels of abstraction. For instance, an easy chair comes under the class of chairs. Up the next level, chairs are furniture. What’s more, the Robo Brain knows that a chair is something you can sit on.