Simultaneous localization and mapping based on multisensor fusion

Abstract

The reliability of simultaneous localization and mapping (SLAM) for mobile robots is low in dense environment due to the high ambiguity of the observed sensor data. A method based on fusion of sonar and vision information is studied to improve the performance of SLAM. Multi-sensor fusion is performed at the level of features, which are extracted from raw sonar data and vision image using Hough transform. Much attention is focused on the full use of redundant information from different sensors in this work. Experimentation with a mobile robot equipped with 16 sonar sensors and a color CCD is carried out, and the results show that multi-sensor fusion is an efficient way to improve the precision and robustness of SLAM.