Development and Evaluation of a Robocentric SLAM Algorithm

Abstract: In this master thesis the front end of a Simultaneous Localization And Mapping(SLAM) system is developed in the programming language C++ and themeta-operating system ROS (Robot Operating System). The algorithms are based onprevious work done at the Swedish defense research agency (FOI) and are a part of aGPS free positioning system developed for military use. The parts that have beenimplemented during the project includes: feature extraction from LIDAR data, featureassociation and a Robocentric Extended Kalman Filter. The sensors used in the SLAMsystem are a Velodyne LIDAR (LIght Detection And Ranging) unit and an IMU (InertialMeasurement Unit). During the master thesis, data collection has been done indifferent types of outdoor environments. The resulting front end SLAM with a Kalmanfilter is evaluated in the different types of environments and compared with bothaccurate RTK (Real Time Kinetic) GPS and a version of the filter that uses datafiltered with a GPS. The GPS free SLAM algorithm in urban and forest environmentsgives position estimates that drifts less than 2% compared with the SLAM algorithmthat has help from a GPS. In open field terrain the GPS free SLAM algorithm hastrouble estimating its position due to a lack of features, which results in significantdrift over time. When the SLAM algorithm with GPS filtered data is compared with anaccurate Real Time Kinetic GPS in an urban environment the average drift is less than1%.