Go on, put your feet up. A new household robot that keeps track of the contents of your kitchen and can learn simple tasks could soon be making you dinner while you relax.The new robot, developed at the Technical University of Munich, exploits the use of radio-frequency identification (RFID) tags on dishes and utensils in its "Assistive Kitchen" to sidestep some of the object-recognition difficulties that have plagued previous household robots.
"If you want to interpret and understand everyday activities using vision data, it's very complicated, error-prone, and resource intensive," says Michael Beetz, who led the research. "If you do it with RFID tags, there is very little sensor information, but it's highly correlated with the activities you are performing."
As a result, the robot knows where everything is, and it can learn simple tasks simply by observing the movements of the objects.
Surfing for recipes
"Setting the table is very easily recognised from cups and plates disappearing from the cupboard and appearing on the table, and cleaning up later is characterised by the same objects disappearing from the table and appearing in the dishwasher," Beetz says.
The team is also working to integrate a number of open-source software packages to enable the robots to get instructions from the internet, in the same way that some search for images .
Robots would search how-to websites like eHow, converting natural language into robot-friendly instructions using language software called WordNet.
They could then optimise the algorithms based on their particular environment, such as carrying four plates to the table rather than making four trips with one plate. And while they are online, they can share what they have learned. Beetz plans a repository of information on which robots can share data about specific tasks, recipes, and handy household tips. "The hard step will be to have the first robot doing it. But then they would share everything," he says.
Intelligent environment
"It's very interesting and promising work," says Stanford University roboticist Andrew Ng. "If you have sensors just on the robot, the range of things the robot can perceive is very limited," he says. "If it is able to use sensors embedded in an intelligent environment, it's as if the robot has many more eyes and sensors and can immediately act much more intelligently in a new environment."
For the robot's core software, Beetz and colleagues chose Player - an open-source software architecture that is becoming a standard for robotics and sensors. Player was pioneered by Brian Gerkey, a co-author of the study. The team plans to make all of its findings available and to publish all of the software that it develops, hopefully to bring you robot dinners that much sooner.
