Perzeptuelle Verankerung ausgedehnter Landmarken während der instruierten Navigation.
Felix Lindner · Informatiktage · 2010
Symbolic artificial intelligence in robotics significantly improves cognitive skills of robotic systems in general. To benfit from the state of the art in artificial intelligence and cognitive science, it must be possible to connect symbolic representations to corresponding sensory information in a complex robotic system. A crucial instance of this problem, i.e. the connection between percepts and symbolic descriptions of objects within an artificial system (referred to as perceptual anchoring), is just heavily discussed in current literature on cognitive robotics. Any agent that follows a route-instruction solves this problem when it identifies the spatial constellations mentioned in the instruction. In this thesis, a framework for perceptual anchoring of extended landmarks based on an analysis of the framework by Coradeschi & Saffiotti (2000) is proposed. The model consists of components to reidentify landmarks, to incrementally reconstruct landmarks’ geometry and to aggregate those geometries to a mental map. The main objective is to construct a experience-based knowledge base that can be used by an agent to reason about and acting in its environment while navigating through. The model will be evaluated in both simulation experiments and within a real-world robot setting.