Robust omniview-based probabilistic self-localization for mobile robots in large maze-like environments

This paper extends our previous work on omniview-based Monte Carlo localization. It presents a number of improvements addressing challenges arising from the characteristics of the given real-world application, the self-localization of a mobile robot in a regularly structured, maze-like and populated operation area, a home store. The contribution of this paper can be summarized as follows: we introduce a more specific extraction of color-based appearance features and propose a novel selective observation comparison method to determine the similarity between expected and actual observation allowing a better handling of severe occlusions or disturbances. Moreover, we present the results of a series of localization experiments studying the impact of the appearance-feature extraction and the observation comparison on the localization accuracy. Our improved approach can successfully demonstrate its omniview-based localization capabilities for a demanding, large operation area - a home store with a size up to 100×60 m². To the best of our knowledge, this is the most complex operation area that has been studied experimentally so far using appearance-based localization techniques.