Robotics researchers understand the increasing roles of robots at home, the workplace, and factories. For this reason, they see the importance of making robots understand how things are done in the real world and how to better understand the humans that they would come in contact with.
As a result, researchers have deduced that teaching robots one skill after another can help them cope with the real world. Some of the skills involve finding one's keys, filling in a glass with something to drink, and having some common sense to decide whether it's right or wrong to cut in the conversation between two people.
These lessons will be stored in Robo Brain which shall act as a central hub. It will be like a huge "encyclopedia" of knowledge that is filled with Internet-gathered information and stored in a robotic user friendly format. This storage of information is readily accessible by robots whenever needed.
"Our laptops and cell phones have access to all the information we want," says Ashutosh Saxena who is an assistant professor of computer science.
Saxena, together with colleagues from several universities such as Brown, Stanford, Cornell, and the University of California, began gathering materials in July. These would include 120,000 downloaded YouTube videos, one billion downloaded images, and 100 million instructional documents and appliance guides. These materials were used hand in hand with the robot training that they have conducted in their laboratories.
Robo Brain shall use images from the materials and process them in order to recognize the objects displayed in them. They should be able to connect images and video with a corresponding text which would enable them to fully recognize the object and understand how it is used. Eventually, they would be able to recognize human behavior and language.
This system of knowing is referred by computer scientists as a type of deep learning in a structured environment. It involves the storing of information in various stages of abstraction.
Robots have computer brain which stores learned lessons in a format known as the Markov model. This type of model is represented graphically with a set of line-connected points.
"The Robo Brain will look like a gigantic, branching graph with abilities for multidimensional queries," says guest researcher Aditya Jami at Cornell who is responsible for designing the extensive database of information for the brain.
On a simpler version, this type of database may be compared to a chart-like representation of how Facebook friends are related to one another. In a scientific sense, it can be akin to the form of a Milky Way.
The project's goal is geared towards the advancement of robotics in order to pave the way for the U.S.' competitiveness in the world economy. Some of its supporters would include Microsoft, Google and Qualcomm.
