Large Scale Sonarray Mapping using Multiple Connected Local Maps

This paper presents a strategy for achieving practical mapping navigation using a wheels driven robot equipped with a sonarray (an advanced sonar array). The original mapping navigation experiment, carried out with the same robot configuration, builds a feature map consisting of commonplace indoor landmarks crucial for localisation, namely planes, corners and edges. The map exhaustively maintains covariance matrices among all features, thus presents a time and memory impediment to practical navigation in large environments. The new local mapping strategy proposed here breaks down a large environment into a topology of local regions, only maintaining the covariance among features in the same local region, and the covariance among local maps. This notion of two hierarchy representation drastically improves the memory and processing time requirements of the original global approach, while preserving the statistical details necessary for an accurate map and prolonged navigation. The new local mapping scheme also extends the endeavour towards reducing error accumulation made in the global mapping strategy by eliminating totally error accumulated between visits to the same part of an environment. This is achieved with a map matching strategy developed exclusively for the advanced sonar sensor employed. The local mapping strategy has been tested in large, real life indoor environments and successful results are reported here.