Simultaneous Localization and Mapping Using Laser Data and Unscented FastSLAM with Scan Matching

Simultaneous localization and mapping (SLAM) is a fundamental problem in autonomous robotic. Many algorithms have been exploited to solve this problem, among these algorithms, FastSLAM is one of the most widely used and Unscented FastSLAM is one of the newest. Although in several scientific researches it is stated that Unscented FastSLAM outperforms FastSLAM, there are still unexamined potentials regarding Unscented FastSLAM. Therefore, this paper seeks to improve the overall performance of Unscented FastSLAM.
Map accuracy and quality directly depend on the accuracy of localization and observations. In SLAM algorithms, robot pose is predicted using motion model, and then corrected using the difference between map features and recently observed features. Accuracy of pose estimation may improve by comparing two sequential observations and modifying robot pose to result in best match between them. This method is called scan matching and has been successfully combined with FastSLAM algorithm and some other SLAM algorithms not including Unscented FastSLAM.
Therefore, this paper seeks to investigate the performance of Unscented FastSLAM combined with scan matching. Simulation results show that combining Unscented FastSLAM with scan match significantly improves accuracy of localization and mapping.