Research On SLAM Algorithm For Robot Based On Stereo Vision

Robot is already widely used in industry, and now it is gradually used in our daily life. The main problem of the popularization of mobile robot is SLAM(simultaneous localization and mapping), which includes many research points, there are many algorithms for solving SLAM, but the research is still going on and the algorithms are improving. The algorithms for solving SLAM are divided into two parts, the filter method based on Bayesian and the optimization method. In all the algorithms based on filter, the first algorithm is Kalman filter(KF). After that, extended Kalman filter(EKF) is proposed to solve the limitation of KF, which is that KF is just suitable for linear system. However, EKF is also not quite suitable for all systems, then the particle filter(PF) is proposed, which can be used for any state or any measurement model in any environment. The only problem of PF is high complexity, and Fast SLAM solved this problem. The algorithms based on Bayesian are already able to solve SLAM, but they are usually limited by the math models they used, the algorithms based on graph optimization used the methods based on global optimization, which use graph model. In the graph model, nodes represent state variables and edges represent constraints between the nodes, the algorithms use the data from sensor to optimize the path of robot, and the map information is based on the path of the robot.In this paper, the algorithms for solving SLAM are implement on ROS, and the simulation environment is Gazebo, the main algorithm is particle filter. This paper also used some other algorithms, the scan matching algorithm based on gradient descent is used to correct robot’s pose, the map building algorithm based on frequency is used to build map. In addition, this paper also used visual sensor to help calculate the changes of the pose. In the experiments, the SLAM process is implemented with all these algorithms, and the main part is the SLAM algorithms and the control of the robot. In addition, in the experiments, the visual sensor is also used to help estimate the moving distance and the rotation angle, in this way, the pose estimation is more accurate, the result and the analysis shows that this method is simple and feasible in some environments.