Mobile Robot Visual-inertial Localization Based On Rotating Magnetometer

With the increasing cost of labor and the rapid development of the manufacturing and logistics industry,mobile robots are playing an increasingly important role in manufacturing and autonomous warehousing.Robot localization based on single sensor has the drawbacks of insufficient accuracy,poor robustness,and great instability in some scenarios,which limits the popularization and application of mobile robots.Integrating information from multiple sensors for robot localization can significantly improve the robustness and accuracy,which has become the mainstream.To improve localization accuracy for mobile robots,this thesis proposes a visual-inertial localization system based on rotating magnetometer.Core technologies of this system include heading angle compensation using rotating magnetometer,visual Simultaneous Localization and Mapping(SLAM)using stereo camera and pose estimation using two-stage extended Kalman filter(EKF).The major contributions of this study are as follows.The visual-inertial localization system is constructed based on a two-stage EKF.A mobile robot using Mecanum wheel is developed for experimental test.Its software is based on ROS,and its hardware includes rotating magnetometer,inertial/magnetic measurement unit,stereo camera and wheel odometer.Magnetic disturbances conceptually fall into two different categories: static disturbances,and spatial disturbances.To remove static magnetic disturbances,direct least square fitting of ellipses is employed for calibration.To remove spatial magnetic disturbances,a method using rotating magnetometer to detect ambient spatial magnetic disturbances is proposed.A criterion named spatial disturbance index(SDI)is defined to characterize the disturbance quantitively.The proposed SDI exhibits good repeatability and has appropriate responses for different degrees of spatial disturbance.An Automatic Heading Reference System(AHRS)using EKF to integrate data from the inertial/magnetic measurement unit(IMMU)is discussed.The quantitative relationship between the SDI and the coefficient of magnetometer measurement error covariance is also established.The EKF measurement error covariance can be dynamically updated in real time to tune the fusion degree of gyroscope and magnetometer automatically according to the SDI,and avoiding the adverse impact of inherent gyroscope drift over time.Eventually,the robot can obtain a relatively reliable heading angle even under strong spatial disturbances,and the heading angle can quickly return to correct value when disturbances disappear.The forward and reverse kinematics model of the omni-directional mobile robot is established,and the wheel odometer based on the four-wheel encoder and robot motion control are realized.The global pose estimation method based on stereo camera is discussed,and the stereo camera models including perspective projection model,projection model for stereo images and distortion model are established.The efficiency and accuracy of various typical feature extraction algorithms are compared and tested under the viewpoint of the actual robot,and the actual performance of the feature-based SLAM system is evaluated.Integrating the heading angle from rotating magnetometer method,the global pose from the stereo camera and the wheel odometer data can therefore yield a visual-inertial system based on EKF.Compared to localization algorithms based on single sensor,the proposed method achieves better localization accuracy and robustness.