Context-based feature extraction with wide-angle sonars

Abstract

The presented context-based approach to feature extraction allows accurate, efficient, and reliable world modelling for mobile robots equipped with wide angle sonars. Densely sampled raw sensor data are clustered online in such a way that the impact of the sonar's angular uncertainty is reduced to a great extent, which allows discrimination between linear and punctual objects as well as accurate determination of their positions relative to the robot. The proposed method works also in partially cluttered environments and it is not limited to a specific sensor configuration or motion planning. The resolution and reliability are achieved by considering the physical properties of the sonars as well as the sequences of sensor states and the corresponding range measurements.