An approach to autonomous navigation based on unscented HybridSLAM

This paper presents a modified version of HybridSLAM (HS) method using unscented Kalman filter to solve simultaneous localization and mapping problem. Instead of applying extended Kalman filter for SLAM (EKF-SLAM) to build the map of the environment, an unscented Kalman filter (UKF) was added to the HS algorithm. This would allow including higher order of non-linearity of the motion. The new method called Unscented HybridSLAM (UHS) is constructing the global map more accurately compare to the original HS and by employing the same constraint of local sub-map fusion technique, a more reliable solution to SLAM problem is achieved. The unscented Kalman filter takes advantage of both statistical and analytical linearization techniques to estimate the global map. The local map in the vicinity of the robot is estimated using FastSLAM and the local map is fused to the map using constrained local sub-map fusion technique. Unscented HybridSLAM uses a minimal set of chosen samples to approximate the posterior mean and covariance for a nonlinear system. The unscented HybridSLAM performance is compared to the original HybridSLAM and FastSLAM algorithms and it is shown that in case of severe nonlinearity, the proposed unscented HybridSLAM is outperforming current filters in terms of estimation of the path and map building.