Representation Learning and Semantics for Robotic Vision

Vision is a powerful tool in enabling intelligent robotic systems to safely and effectively perform useful tasks. An important precursor to well-considered robotic action is a semantic understanding of the environment, which can be facilitated by machine learning techniques. Recent advances in the field of machine learning for computer vision seldom make considerations for robotic applications. In this thesis, important problem domains for robotic vision are identified and explored. Machine learning methodologies are developed with motivation drawn from the needs of a robotic vision system.