EKF Localization and Mapping by Using Consistent Sonar Feature with Given Minimum Landmarks

The SLAM or localization needs successful data association of the detected feature with landmarks. Well described features of the environment are essential for good data association. In this paper, the localization of the robot is executed by the extended Kalman filter (EKF) with given minimum landmarks of the environment. Consistent features for localization are extracted by using only sparse sonar data. Features are extracted by using a sonar data clustering from a footprint-association (FPA) method and a feature fitting from a least squares (LS) method to overcome challenges associated with sonar sensors, such as a wide beam aperture and a specular reflection effect. The extracted features are, also, evaluated as a post-processing through the probabilistic association which associates the extracted feature with the weighted average probability of the grids that are located within the area of position uncertainty of the feature. The proposed methods have been tested in a real home environment with a mobile robot