Researchers create cloud-based brain for robots

A group of European researchers has released the first version of a cloud computing platform for robots that will help them take advantage of powerful virtual resources. Essentially, they’re treating robots like any other device — desktop, tablet or mobile phone — running web applications, only robots can learn from each other and can do a lot more than just update screen displays.
The project, carried out by a team at ETH Zurich, is called RoboEarth and its linchpin is a cloud software platform called Rapyuta. The way it works is pretty simple at a high level: robots communicate with a cloud-based application platform that carries out computation tasks and connects to a cloud database full of information such as maps, images, language, as well as to other web services. The robots themselves are pretty much hardware terminals equipped with sensors and moving parts but limited on-board processing power or data storage.
This, of course, is an idea that has been with us since the mainframe computer and continues today via cloud computing and web and mobile applications. Why limit a device to its own physical capabilities when there’s an infinite (although, in the case of mainframes, not so much) expanse of computing power, memory, storage and data available in the ether? As long as the device has a strong internet connection, it doesn’t need a massive hard drive or the latest, greatest processor.
re_architectureOne thing RoboEarth does a little differently, though, is allow for databases that robots can update as they go about their business in different situations in different parts of the world. It’s machine learning, only in a much more literal sense: robots are actually learn from the experiences of other robots. That learning, in turn, should make them more useful to humans who won’t have to program them as thoroughly and perhaps can use the robots to perform a wider — and ever-expanding — variety of tasks.
Presumably, though, some statistical machine learning on the backend could make the robots even smarter as they generate more and more data and patterns begin taking shape. (We’ll be talking about unique ways to put machine learning to work at our Structure: Data conference next week in New York.)
The easy joke to make about this type of project is to say it’s the start of SkyNet and the rise of the machines, but that’s a bit of a stretch. After all, the machines themselves aren’t communicating with one another but, rather, with a centralized computing infrastructure operated by humans. It’s similar to IBM’s Watson system, which is really good at answering questions, but only as good as its information database and algorithms allow it to be.
If you’re curious to learn more about the promise and limitations of something like RoboEarth, Markus Waibel, one of the project’s researchers, has a great blog post explaining his vision of the project and where it fits into the greater ecosystem of web-based robotics.
Feature image courtesy of Shutterstock user Bruce Rolff.