Robust SVSF-SLAM Algorithm for Unmanned Vehicle in Dynamic Environment

摘要

The researches of Simultaneous Localization and Mapping (SLAM) are very important problems for mobile robot in dynamic environments. This paper shows a new approach joining data given by an odometer and a laser extend discoverer sensor to proficiently explain the SLAM problem of Unmanned Ground Vehicles (UGV). Our approach to self-localization in a non-static environment uses an important step in the map management algorithm that is robust to delete the moving feature. The robustness is accomplished by taking a part of the dynamic object instead of all features at once, meaning that partial occlusion will only affect a subset of all features. Our hypothesis is that a feature that is moving with an important velocity is considered that is out of date and it cannot maintain much useful information. The updates of dynamic landmarks are calculated every time step, as well as a measurement update which is the dynamic object positions become more uncertain as they move. A new Adaptive Smooth Variable Structure Filter (ASVSF) SLAM algorithm is implemented to localize the UGV with an original covariance matrix formulation. The proposed algorithm is validated in real-world and the results obtained confirm the efficiency and robustness in a dynamic environment.