Underwater Simultaneous Localization and Mapping Based on 2D-SLAM Framework

The location problem of underwater robot is always the key of its underwater operation. At present, most underwater robots use inertial navigation and dead reckoning to locate. The positioning error can be corrected by means of GPS or acoustic underwater beacon, but these operations are not suitable for a large area of unfamiliar waters. The Simultaneous Localization and Mapping (SLAM) was extensively used in terrestrial robots. If applying it to underwater ones, the underwater robots can build a map with its own environmental through sensing sensors and correct positioning errors autonomously. In this paper, the mechanically scanned imaging sonar (MSIS) data from the submarine cave were collected for simultaneous localization and mapping, by using the existing mature laser radar 2D-SLAM frame. The main works are that we use pre-processing operations including beam segmentation, noise removal and threshold filtering to make sonar scan lase-like. After these operations, the sonar data can be transformed into a 2D laser radar packaging format and use in the frameworks. By using odometer information and IMU data in the data set, the motion distortion of the sonar point cloud is removed. Then we improve the Gmapping frames to make it more suitable to this underwater data set. Finally, the sonar-based dataset is successfully applied to the classical 2D laser SLAM frames Gmapping and Cartographer.