Until recently, robots have not been capable of understanding and coping with unstructured environments (like the ones humans work in) because their systems have relied on knowing in advance the specifics of every possible situation they might encounter. Each response to a contingency has had to be programmed in advance, and systems have had to rebuild their world model from sensor data each time they had to perform a new task.
This is one of the main reasons why, to date, robots have been mostly relegated to highly controlled and predictable environments like manufacturing plants, but have made few significant inroads into the human sphere. The human world is just too nuanced and too complicated to be summarized within a limited set of specifications.
But what if robots could learn from their past experiences? And what if they could share their new-found knowledge instantaneously with their peers?
These are not hypothetical questions. 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. Researchers are already using networked 'feed forward' approaches to make significant advances in machine-based learning systems.
Thus far, however, these smart feed forward systems have been operating in isolation from each other. If they are decommissioned, all that learning is lost. Even more disconcerting to researchers is the question: why are thousands of systems solving the same essential problems over and over again anyway?
The aim of RoboEarth is to use the Internet to create a giant open source network database that can be accessed and continually updated by robots around the world. With knowledge shared via the cloud on such a vast scale, and with businesses and academics contributing independently on a common language platform, RoboEarth has the potential to provide a powerful feed forward to any robot's 3D sensing, acting and learning capabilities.