A network and repository of data, where robots can share and learn from each other about the tasks they perform, has been demonstrated at the Eindhoven University of Technology (TU/e). The project is the culmination of four years of research by scientists from TU/e, Philips, ETH Zürich, TU München and the universities of Zaragoza and Stuttgart.
“The problem right now is that robots are often developed specifically for one task”, says René van de Molengraft, TU/e researcher and RoboEarth project leader. “Everyday changes that happen all the time in our environment make all the programmed actions unusable. But RoboEarth simply lets robots learn new tasks and situations from each other.”
The system also allows computing to be carried out using its cloud engine, meaning that robots need not necessarily have the required processing power or store the information themselves. Data stored in RoboEarth includes software components, maps for navigation, task instructions and object recognition models.
A simulated hospital setting was used as a means to demonstrate RoboEarth. Initially, two robots explored and mapped two hospital rooms using instructions stored in the RoboEarth cloud. The same set of instructions was able to be used, despite the robots being different models. A third robot was shown being asked for a drink by a hospital patient, and referring to RoboEarth to find out how to complete the task. A final robot without prior knowledge of the room delivered the drink using the maps created by the first two robots and hosted in the cloud.
The demonstration marks the end of the RoboEarth research project, but having been developed in an open-source manner, the system can still be used and improved upon by external users.