The scenario of the computer/machine revolution is as old as the computers themselves, and an entire generation of Science Fiction movies, books and games has been built on top of it. The Robo Brain program threatens to make the craziest -and dangerous- dreams of every FE fan a reality, as its focus is the gathering of skills and knowledge from publicly available information sources -in short, the Internet- in order to «feed» the robots with knowledge.
«As mentioned on the program's website, Robo Brain is a large-scale computing system that «learns from publicly available sources on the Internet, computer simulations, and robot tests in the real world. It gathers anything related to robotics into an extensive and interconnected knowledge base». As noted, its applications include creating prototypes for robotics research, domestic robots and autonomous vehicles.».
A similar project is being developed by researchers at Eindhoven University in the Netherlands: RoboEarth, which has been described as “a world wide web for robots,” is intended to be a global database that robots can access—however, Robo Brain will obtain the required information (which it will make available to “interested” robots for daily tasks) via the Internet instead of requiring special programming from humans.
The project involves researchers from Cornell, Brown, Stanford, and UC Berkeley, and is supported by technology companies such as Google and Microsoft.
The system began processing information from the Internet last month, with its researchers saying it can handle about a billion images, 120,000 YouTube videos and 100 million instruction and manual files. According to the researchers, Robo Brain is not only capable of recognizing objects, but also of understanding how they are used - as well as more complex concepts, such as human language and behavior.
“Our laptops and mobile phones have access to all the information we want. If a robot is faced with a situation it has not seen before, it can search Robo Brain in the cloud,” explains Cornell University’s Asutos Saxena, an assistant professor of computer science.
The Robo Brain processes images to distinguish the objects within them, and by linking images and videos with text, it learns to recognize what they are and how they are used – as well as, as mentioned earlier, human language and behavior. The system uses what computer scientists call «structured deep learning», where information is stored at multiple levels of abstraction: a chair is classified as a type of chairs, and chairs are furniture. Someone can sit on a chair, but can also sit on a step, etc.
«The Robo Brain will resemble a gigantic, multi‑branched graph with capabilities for queries in many dimensions» notes Antitia Zami, the researcher who designed the large‑scale brain database—specifically, it can refer to a graph that depicts friendships/relationships on Facebook, but on a scale that probably corresponds to the Galaxy. Meanwhile, the Robo Brain will also have «teachers», thanks to crowdsourcing: on the program’s website the items the system has «learned» will be displayed and visitors will be able to make additions and corrections.
The program is supported by the National Science Foundation, the Office of Naval Research, the Army Research Office, Google, Microsoft, Qualcomm, the Alfred P. Sloan Foundation and the National Robotics Initiative.
Source: koutipandoras.gr
