Square Root Unscented Particle Filtering for Grid Mapping

In robotics, a key problem is for a robot to explore its environment and use the information gathered by its sensors to jointly produce a map of its environment, together with an estimate of its position: so-called SLAM (Simultaneous Localization and Mapping) [13]. Various filtering methods – Particle Filtering, and derived Kalman Filter methods (Extended, Unscented) – have been applied successfully to SLAM. We present a new algorithm that applies the Square Root Unscented Transformation [14], previously only applied to feature based maps [7], to particle filtering for grid mapping. Experimental results show improved computational performance on more complex grid maps compared to a well-known existing grid based particle filtering algorithm, GMapping [2].