Autonomous vehicles usually use more than one positioning system to improve
their position estimate. Some positioning systems are advantageous in cert
ain types of environments, while others are more efficient in others. This
paper describes a data fusion method, where the differences between measure
ments are used to identify the type of terrain through which the vehicle is
traveling. In this system, position estimate by odometry is compared to th
at calculated by triangulation, and the differences are fed into a neural n
etwork. This neural network, which is pertained by a set of different terra
in types, classifies the examined environment by matching it with the most
similar environment it can "recognize". (C) 1998 Elsevier Science Ltd. All
rights reserved.