RBF network for mobile robot sonar based localization and environment modeling

Abstract

In the paper an RBF-ANN (Radial Basis Function Artificial Neural Network) based localization of a mobile robot and environment modeling is proposed. An ultrasonic distance measurement unit is used for scanning the mobile robot's environment. The environment is learned by the RBF-ANN and the result is compressed in the network weights. To each specific position/orientation of the robot corresponds a set of attached weights which are stored in a database. With a convolution/correlation operation the network weights corresponding to an actual position of the mobile robot are compared with the stored weights from the database. The best correspondence determines the possible position of the robot. In the paper the result validation is presented.