Simultaneous Localization and Mapping with One Particle

Rao-Blackwellized particle filter (RBPF) is one of the powerful techniques for simultaneous localization and mapping (SLAM) since it factorizes the posteriors over the maps and potential trajectories of the robot. RBPF has its own advantages as well as disadvantages. RBPF approach has a major problem with computational complexity and this is related to the number of the particles. Thus, reducing the number of particles is one of the biggest challenges in RBPF. In this study, one particle based on the proposal distribution is computed by utilizing the most likely pose obtained by a scan matching procedure. Next, grid search around the most likely pose is performed to search for better pose. Furthermore, the proposal distribution does not depend on the odometry model and formulation of computing the probability of the sensor model is given. After that, mapping is built based on the computed pose. Lastly, accuracy-based quantitative and qualitative results by utilizing 2D LIDAR performed in computer simulations show the efficacy of this method.