By Andy Choi
The robot Internet, called RoboEarth, is a groundbreaking project that allows robots to share information with one another.
As robots carry out tasks and compile feedback about them, they upload what they've learned or relevant data about the task or the immediate environment to their worldwide web. That information then becomes available to other robots, which can use it to improve their own functions and performance. In short, the network enables robots to take part in a massively collaborative and ongoing learning process.
The type of data being shared in this repository include images and object models to enable object recognition, maps and world models to aid navigation, manipulation strategies for completing tasks that require dexterity, and other bits of information that contribute to artificial intelligence.
Fears of robot mutiny aside, this increased intelligence could rescue robots from the controlled and predictable environments (like manufacturing plants) where they are currently used the most and put them to work in more fluid human spaces, like hospitals and schools.
The team behind the project comprising researchers from Eindhoven University of Technology, Swiss Federal Institute of Technology Zurich, Philips, and University of Zaragoza, has received multi-year funding from the European Commission's Cognitive Systems and Robotics Initiative.
“We started RoboEarth partially out of a frustration with the fact that there are so many robots out there that do repetitive tasks, but they never improve,” RoboEarth program manager Markus Waibel from the Swiss Federal Institute of Technology in Zurich told the Star.
“In robotics there is a lot of reinventing the wheel going on,” he added. “Most current robots see the world their own way, and there’s very little standardization.”
Basically, robots need a structured environment to work. Each response to any situation has to be programmed in advance and accessed by the robot as it encounters new sensory information.
“Robots have been mostly relegated to highly controlled and predictable environments like manufacturing plants, but have made few significant inroads into the human sphere,” reads the RoboEarth website. “The human world is just too nuanced and too complicated to be summarized within a limited set of specifications.”
“Rapid development of sensor and networking technology is now enabling researchers to collect vast amounts of sensor data, and new data-mining tools are being developed to extract meaningful patterns,” reads the RoboEarth website.
But until RoboEarth, there was no single system connecting all of this data.The minds behind the website envision a platform such as Wikipedia for our circuit-board laden friends to share and learn from:
“Wikipedia is something people use very successfully to share knowledge, where everyone can benefit from the knowledge that people have,” said Waibel. “There isn’t something like that for robots.”
If you asked a robot to get you a glass of water, the chances of it rushing toward the cupboard for a glass are as likely as it zooming in the direction of the toilet, Waibel points out. That’s because robots have no preconceived notion of how objects fit into the world.
“What is a glass of water? Where should I start looking for the glass of water? These are all questions the robot needs to answer,” says Waibel. “We have found a way to represent this information so that different robots can understand it and it is meaningful.”
The key is allowing robots to share knowledge, he adds.
The makers of RoboEarth hope the technology will help the growing number of domestic robots that we are expecting in our homes over the next few years.
A service robot would be able to tap into a constantly updated database of house maps and sensory data, so if it needed to set the table, it could enter the kitchen, find the knives and forks, and interact with them properly, even if it had never done the task before.
An open source version of RoboEarth is scheduled for release in July 2011.