Elastic nets and vision-based autonomous navigation

Navigation is an essential task in autonomous vehicle guidance. Basically, two different approaches for mapping the environment have been proposed. Geometric maps model the world in one rigid coordinate system yielding the problem that precise global measurements become necessary to establish a map. Topological maps rely only on local measurements and often suffice to guide a vehicle on its path. The author discusses the interrelation of local measurements and global ones. Local measurements can be obtained by ´internal´ sensors like odometers or gyroscopes. Their error behavior is modelled by a systematical differential error and a random error portion. Often, global measurements are obtained optically (e.g. by laser triangulation or a landmark recognition). The point is that these systems reference directly to fixed external positions.