Mobile robot self-localization from range data using view-invariant regions

The problem of mobile robot self-localization given a polygonal map and a set of range readings from a ring of sonar sensors is addressed. It is assumed that straight-line edge segments can be extracted from the contour defined by the sensed range data. Previous approaches used a tree search complemented by a number of good pruning heuristics to match observations to map edges. The approach here is to modify this search by decomposing the map into view-invariant regions and searching only for the edges that are visible from a point in the constructed regions. The view-invariant regions are computed offline as part of map construction. The goal of this map preprocessing is to better organize the nodes in the search tree in order to improve the efficiency of the search task.<
>