Stereo vision as a sensor for SLAM based Smooth Variable Structure Filter with an adaptive Boundary Layer Width

The autonomous navigation task of a mobile robot depends on its ability of localization and owning a description about its environment. To deal with these requirements, robots need to be equipped with Simultaneous Localization and Mapping (SLAM) module. This earlier could be solved by many approaches, mostly based on the stochastic approach, using extended Kalman filter (EKF) or the Rao-Blackwellized particle filter. The SLAM has been already approached using a new alternative filter which is the Smooth Variable Structure Filter (SVSF). This estimator is a predictor corrector formulated on the theory of sliding mode control and variable structure systems. The first version of SVSF uses a predefined Boundary Layer Width vector and don't require covariance derivation. In this work, we propose using a new form of SVSF to deal with the SLAM problem based on an adaptive (optimal) boundary layer matrix. The (ASVSF) is very robust estimator against modeling errors and noises and keeps a compromise between robustness and accuracy. Visual SVSF-SLAM and ASVSF-SLAM are implemented, validated with experimentation and compared with EKF-SLAM algorithm. The comparison of simulation results proofs the efficiency, robustness and the accuracy of ASVSF-SLAM comparing to the other algorithms, while the experimental results show that ASVSF-SLAM is the less accurate.