Imagine for a second that there’s a huge online database that is compiling images, videos, and instructions from across the web. The database translates all of that information into a robot friendly code that robots can access and download. So, hypothetically a robot who has never been programmed to turn on a light could visually scan a light switch, match that image with one in the database, download the information attached to that image, and know how to flip the switch. You don’t actually have to imagine any of that because it’s something that exists in reality, and that database is called Robo Brain.
Robo Brain is a huge step in cloud robotics. Robots will be able to learn tasks they weren’t programmed to do by accessing coded instructions in the cloud. This is an essential technology when it comes to intuitive or learning robots. It also allows for robots to become more specialized.
The Robo Brain project went online in July and researchers from Cornell, Brown, Stanford, and the University of California, Berkeley are downloading billions of images, hundreds of thousands of videos, hundreds of millions of how-to’s and manuals, to the database.
The system makes use of multiple levels information that are stored in abstraction that scientists call “structured deep learning”. It’s all about identification and making connections. An electric guitar is a subset of guitar, a guitar is a subset of musical instruments. People can play guitars and also the piano which is another musical instrument. Pianos have keys and so do computer keyboards. People also use computer keyboards.
It will be interesting to see whether or not Robo Brain will be successful in training robots to perform unprogrammed functions, and whether or not it can translate to manufacturing. Streamlined usually means optimized, but streamlined would be much more efficient if it were specialized. That’s what Robo Brain could offer manufacturing.