Autonomous mobile robots in household environments have to cope with many different kinds of objects which they must detect, recognize, and manipulate. Over their lifetime, the robots must adapt to new objects and incorporate new perception methods. In this paper we present a system for life-long learning of training data and perception method parameters using a document-oriented, schema-less database technology that is typically used in cloud computing applications. Not only can a single robot extend and increase its data volume continuously over time, but it can also potentially share this very dataset with multiple other robots through the cloud.