Addressing perceptual aliasing in topological map-building: economical use of metric information in a cognitive framework

Two or more different entities that appear identical when observed in isolation are perceptually aliased. While the nature of perceptual aliasing is clearly general and its occurrence depends on the kind of entities and perceptual skills assumed, it originates a very practical problem when it comes to acquiring purposeful spatial representations, e.g., maps, of an unknown environment under exploration. If perceptual aliasing is taken into account, whenever a place is perceived as identical to one or more encountered earlier in the exploration, the following question must be addressed: “Is this place one of those visited earlier, or a totally new place (Was this a loop)?”. Different approaches to map-building handle (if that) this issue in a variety of ways. Our work takes place in the framework of the Spatial Semantic Hierarchy, which has proved relevant to a diverse body of research in AI, robotics and cognitive sciences. It represents spatial knowledge over different overlaying ontological levels, and allows topological map-building. In its current computational implementations, the complete set of topological hypotheses originated by perceptual aliasing is pruned by further exploration, and enforcing topological and common sense properties. Many such hypotheses, however, may survive due to symmetry in the environment or while the exploration is not complete, yielding a combinatorial explosion of possible maps. We believe that taming this proliferation is possible without giving up the complete account of all the possible hypotheses, by ranking them according to their inherent metrical plausibility. We propose to evaluate such a parameter from local metrical annotations, in a computationally economical fashion, and without building a single global metrical frame of reference, that would introduce the hard problem of enforcing global metrical consistency. The main contribution here is the novel perspective, purpose and way mathematical methods and spatial ontologies are employed and integrated.