Research On Precise Localization Technology Of Wheeled Mobile Robots Based On Multi-Sensor Information Fusion

With the development of science and technology and people’s pursuit of quality of life,various types of indoor robots have been integrated into human daily life to liberate human hands and increase efficiency,among which wheeled mobile robots are widely used.The precise positioning of wheeled mobile robots is the basis for completing other tasks.Sensors commonly used to achieve positioning include 2D laser radar,wheel odometry,and Inertial Measurement Unit(IMU).Single positioning sensors are difficult to provide accurate positioning information.Research on the precise positioning of wheeled mobile robots through multi-sensor information fusion is of great significance for the application of wheeled mobile robots.The main research content of this thesis is as follows:Firstly,for indoor environments,the hardware for precise positioning of wheeled mobile robots is selected,and a wheeled mobile robot experimental platform with 2D laser radar,wheel odometry,IMU,and Mecanum wheel motion chassis is built.The motion model and wheel odometry model are derived.The 3D model of the wheeled mobile robot is established using Solidworks software,and its urdf file is exported using a plugin.The software platform of the wheeled mobile robot is built based on the ROS(Robot Operating System)platform.Secondly,to solve the problem of inaccurate positioning caused by wheel slippage and idling,the Extended Kalman Filter(EKF)algorithm is used to fuse wheel odometry and IMU data information based on the motion model and wheel odometry model of the wheeled mobile robot,improving the positioning accuracy of the wheeled mobile robot.Suitable covariance matrices for wheel odometry and IMU are set on indoor tile floors,and the accuracy of multi-sensor information fusion positioning using the EKF algorithm is verified through experiments.Finally,to address the problem that the EKF algorithm is easily affected by external environments and requires readjustment of positioning sensors after the wheeled mobile robot changes working environments to achieve higher positioning accuracy,the Gate Recurrent Unit(GRU)neural network algorithm is used for multi-sensor information fusion positioning.After training with sensor data collected in different ground environments,the algorithm is deployed on the wheeled mobile robot,and wheel odometer and IMU data information are fused.Through experiments,it is verified that the GRU neural network fusion positioning has high positioning accuracy under different ground conditions.Experimental results show that the accuracy of EFK fusion positioning using covariance matrices for rubber-hardened ground on tile floors and for tile floors on rubber-hardened ground is slightly lower than that of GRU neural network fusion positioning.In the path planning experiment,after setting the endpoint coordinates,the average error of GRU fusion positioning is 0.095 meters,which is smaller than the average positioning error of the original wheel odometry of 0.184 meters,after global path planning algorithm A*algorithm and local path planning algorithm TEB algorithm are used for path planning.Therefore,the GRU neural network multi-sensor information fusion positioning used in this thesis can improve the positioning accuracy of wheeled mobile robots and still maintain high positioning accuracy after changing working environments.