Voronoi Graph Matching for Robot Localization and Mapping

摘要

In this article, we develop a localization and map building approach for a mobile robot in which annotated generalized Voronoi graphs are used as the robot's spatial representation of the environment. The core of our approach is a matching scheme for solving the data association problem of identifying corresponding parts in two tree-formed Voronoi graphs, one representing a local observation and the other one representing the robot's internal map. Our approach is based on the notion of edit distance which means it computes the cost optimal way to transform both graphs into the same graph. The costs for adding or deleting branches are based on a measure that assesses the stability of nodes and edges in the graphs as well as their relevance for navigation. In addition, we incorporate spatial constraints based on the graph annotations which leads to a top-down dynamic programming implementation that performs reliably and efficiently in practice.