Abstraction, ontology and task-guidance for visual perception in robots

ion, ontology and task-guidance for visual perception in robots Matthias J. Schlemmer and Markus Vincze Vienna University of Technology, Automation and Control Institute 1040, Vienna, Gusshausstrasse 27-29/E376, Austria [email protected] Abstract. In this paper we summarise theoretical work on visual perception for cognitive robotics. We study the use of logic-based ontologies for situated perception, motivated by cognitive science and philosophy. General considerations on scope and aims of “cognitive vision” is given where a functional position is taken. The use of symbol-based reasoning following this approach and the central importance of the mapping from low-level sensor data to higher-level symbols is outlined. Special focus is given w.r.t. the localisation of different mechanisms (subsymbolic, symbolic) during perception and cognition and cognitive functions that seem important are named. A preliminary implementation approach for mapping simple Gestalts to object concepts and a general discussion on the use of logic for perception conclude the paper. In this paper we summarise theoretical work on visual perception for cognitive robotics. We study the use of logic-based ontologies for situated perception, motivated by cognitive science and philosophy. General considerations on scope and aims of “cognitive vision” is given where a functional position is taken. The use of symbol-based reasoning following this approach and the central importance of the mapping from low-level sensor data to higher-level symbols is outlined. Special focus is given w.r.t. the localisation of different mechanisms (subsymbolic, symbolic) during perception and cognition and cognitive functions that seem important are named. A preliminary implementation approach for mapping simple Gestalts to object concepts and a general discussion on the use of logic for perception conclude the paper.