Statistical approach to integration and interpretation of robot sensor data

Making use of multiple sensors is crucial for improvement of mobile robot navigation performance. The article introduces a problem of integrating noisy range data by data fusion on signal/pixel level. This is performed for multiple sensors and different sensor positions into a common description of the environment. The article deals with grid models, which are used as low level representations of sonar range measurements as well as for data fusion process. A short overview of used methods for the grid representation is shown and different fusion methods for various sonar models are discussed. Special attention is paid to the influence of sonar modeling on robustness of the data integration process. The article deals with methods for improvement of fusion robustness. Novelty of the presented approach stands in a sonar model which is designed as dependent on measured distance. The other improvement is an optimal feature selection for doorway recognition. The designed methods are illustrated by examples and test runs on an experimental mobile platform with a sonar sensor system.