Think of it: an intelligent network comprised of autonomous robots. No, not like the Terminator's self-aware computer system bent on destroying humanity, but a Wikipedia-like resource for robots to learn, and to share their newly found knowledge. It's called RoboEarth and it's being built by Swiss researchers.
If robots are ever to become useful, they have a lot to learn. "Today's robots don't know many of the things we take for granted: milk turns bad if it's not in the fridge. The milk bottle can break. The oven gets hot when it's turned on. We learn these and many more things about our world in our first years of life. But, although they seem so basic, today no robot has this basic knowledge," said Markus Waibel, from the Swiss Federal Institute of Technology(ETH)in Zurich.
RoboEarth will enable robots connected via its network to share information, learn and increase their overall memory. "It is not intended as a communication network. Rather, it will construct a knowledge base, much like Wikipedia," Waibel said.
The perception of artificial intelligence has changed over the years, Waibel said.
"Initially we thought logic was the highest form of intelligence. Today we know that there is much more to building an artificial intelligence than logical reasoning.
"A network like RoboEarth will likely exacerbate current ethical, moral, and legal challenges. However, for now these issues are dwarfed by the technological challenges."
The reason is that the more varied robots are the more difficult sharing knowledge between them becomes. "We humans excel at dealing with unstructured information, however, for a robot even a fairly structured knowledge resource like Wikipedia is too ambiguous and incomplete. For example, Wikipedia lists 23 alternative meanings for the word robot," Waibel said.
Structuring data categories to help robots learn faster
"Ultimately, an Internet for robots will have to cover a lot more knowledge than Wikipedia and need to be far more structured." Waible and his colleagues RoboEarth scientists are currently using KnowRob, a taxonomic structure to help robots learn faster. This will help robots learn about data categories that include everything from geography and basic physics to manipulation tasks and offline learning.
Currently, the European Commission FP7 funded project is using a single server to host the RoboEarth database. "However, the software architecture is built on a cloud computing framework that is highly scalable," said Waibel.
RoboEarth's scientists want robotic systems to benefit from the experience of other robots in other environments, paving the way for rapid advances in machine cognition, behavior, and eventually, better human and machine interaction. "While science fiction writers have imagined artificial intelligences in Terminator and Space Odyssey series, I think those analogies are flawed. RoboEarth is about building a knowledge base, and while it may include intelligent web services or a robot app store, it will probably be about as self-aware as Wikipedia," said Waibel.
The release of an open source version of RoboEarth is scheduled for July 2011.