Managing Belief States for Service Robots

Abstract

In this thesis, the powerful paradigm of Unstructured Information Management is applied to the problem of robot perception. We present RoboSherlock, an open source software framework demonstrating the vast potential of the paradigm for real-world scene perception. A large number of existing and novel methods, from classical computer vision to probabilistic reasoning, are integrated into a highly flexible, extensible and capable system generating object descriptions of unparalleled information richness.