Ambiguous Landmark Problems in Cognitive Robotics: A Benchmark for Qualitative Position Calculi.

摘要

In this paper we introduce a task which can serve as a bench-mark for qualitative relative position calculi. In this task am-biguous local landmark observations have to be integrated into survey knowledge. We show that the most prominent rel-ative position calculus, Freksa’s Double Cross Calculus can solve a specific instance of this task. The observations can be represented in a constraint network and standard constraint propagation solves the ambiguity problem. However, more general instances of the ambiguous landmark problem cannot be solved using the Double Cross Calculus. Therefore we present an extension to relative position ternary point configuration calculi which uses an adaptable level of granularity. This family of calculi is capable to solve general instances of the proposed benchmark. Thereby robot applica-tions including reasoning about ambiguous perceptions will be made possible.