Research And Application Of Robot Localization Based On Multi-sensor Fusion

With the gradual deepening of the application of indoor robots,the robots autonomous navigation ability has become one of the bottlenecks that limits the improvement of the service ability of indoor robots.Among them,localization is the foundation and basis of autonomous navigation and is one of the core technologies that restrict the wide application of indoor robots.Since it is difficult to ensure the robot to work stably in various indoor environments for a long time with the localization of single sensor,the fusion of multiple sensors' data for localization has been extensively studied and applied.Based on the analysis of localization with single sensor,this paper proposes the fusion algorithm with monocular vision,IMU and wheel odometer.Firstly,the relevant theories of monocular vision localization and IMU localization are respectively studied,and the monocular ORB-SLAM algorithm is discussed,which lays the foundation for multi-sensor fusion.According to the relevant experimental analysis,it is concluded that the monocular visual localization is less robust and the IMU localization error is serious.Aiming at the problems of single sensor localization,this paper proposes a fusion algorithm with monocular vision and IMU based on ORB-SLAM algorithm.On the basis of monocular ORB-SLAM algorithm,the fusion algorithm improves Tracking,Local Mapping and Loop Closing threads,and adds data preprocessing threads and initialization threads.The experimental results in the EuRoc dataset show that the fusion algorithm with monocular vision and IMU has the advantages of higher localization accuracy and better localization robustness compared with the monocular ORB-SLAM algorithm.In addition,the fusion algorithm with monocular vision and IMU in this paper has slightly less localization robustness but better localization accuracy compared with the excellent monocular VINS-mono algorithm in the current fusion algorithm,which is suitable for indoor robot localization.In view of the problem that the fusion algorithm with monocular vision and IMU does not consider the zero deviation of accelerometer sensor and the error of gravity component estimation,this paper introduces a wheeled odometer sensor into the fusion algorithm with monocular vision and IMU,and proposes a fusion algorithm with monocular vision,IMU and wheel odometer.In the fusion algorithm,IMU and wheel odometer data are fused in a joint pre-integration method,and ORB algorithm is improved by using the joint pre-integration.Besides,the objective function is constructed by using the joint pre-integration error and visual reprojection error.In order to evaluate the effectiveness of the fusion algorithm with monocular vision,IMU and wheel odometer,this paper uses Turtlebot2 robot to carry out relevant experiments.The results show that the fusion algorithm with monocular vision and IMU and the fusion algorithm with monocular vision,IMU and wheel odometer can effectively improve the localization problems of single sensor.In addition,the fusion algorithm with monocular vision,IMU and wheel odometer has higher localization accuracy,better localization stability and excellent performance in indoor ground environment compared with the fusion algorithm with monocular vision and IMU.