The Implementation Of Vision Based Unscented FastSLAM Algorithm For Mobile Robot

With the fast development of science and technology, the automation in industry isbeing improved gradually. Big development has been made in mobile robotics since itinvolves automation, computer science, artificial intelligence and so on. SimultaneousLocalization and Mapping(SLAM) has always been the focus of the robot navigationfor many decades, SLAM is the basic and necessary factor for mobile robot to realizeautomation and intelligent. EKF-SLAM algorithm and FastSLAM algorithm based onparticle filter are two popular algorithms for SLAM, which have been widely used inrobot navigation for many years.Since the EKF-SLAM has two famous ?aws, i.e. high computational complex-ity and sensitivity to the wrong date association, it can not be suitable for large scaleenvironment. This thesis proposes an Unscented FastSLAM algorithm based on Rao-Blackwellized particle filter and Unscented Kalman filter(UKF). Unscented FastSLAMalgorithm is improved based on FastSLAM2.0 algorithm, by using UKF instead of thelinear approximations of the nonlinear function, and using the effective number of par-ticles as a criteria to reduce the particle degeneration. Simulations and experiments areperformed to demonstrate that the Unscented FastSLAM algorithm performing much bet-ter than FastSLAM algorithm on the accuracy and the robustness especially when themeasurement noise is bigger.