Research On Robot Localization Algorithm Based On Multi-Sensor Fusion

In recent years,robots have begun to emerge in all walks of life.The robot localization algorithm based on multi-sensor fusion is the basis for robots to realize autonomous movement and is also one of the core technologies of robots.Multi-sensor fusion localization fuses the observation information of multiple sensors to provide accurate and stable localization information for the robot.However,the environment in which the robot runs is complex and diverse,and in certain specific scenarios,the existing algorithms cannot meet the localization requirements.Therefore,this article focuses on the localization problem in the environment with completely symmetrical geometric shapes and the scene with missing visual features.The AMCL(Adaptive Monte Carlo Localization)algorithm fused with visual mask information and the stereo and IMU(Inertial Measurement Unit)tightly coupled localization algorithm are studied separately.The main research content of this article is as follows:(1)Building sensor model and calibrateing sensors.Observation models are constructed for various sensors such as lidar,vision camera and IMU,and different methods are used to calibrate the monocular camera,stereo camera and IMU to obtain the internal parameters of the sensors,and joint calibration to obtain the external parameters between the sensors.(2)Local pose correction based on monocular vision.In some specific tasks,the robot is required to reach the specified target point accurately.The accuracy and stability of the pose estimated by the AMCL algorithm alone cannot be guaranteed.This paper proposes a local pose correction algorithm based on monocular vision.(3)In an environment where the geometry is completely symmetrical,the robot localization algorithm based on AMCL is prone to pose estimation symmetry problems,and the localization algorithm cannot detect the occurrence of symmetry problems.In order to solve the problem of AMCL localization failure in a symmetrical environment,this paper proposes an AMCL localization algorithm fused with visual mask information.By adding visual masks to the known map,an environmental map with mask information is constructed,and the monocular camera is used to detect and recognize the visual mask,and the pose information of the robot is calculated.Combine AMCL and monocular vision pose to solve the problem of robot kidnapping.(4)Pose Estimation in Vision Loss Scene.Based on the visual SLAM(Simultaneous Localization and Mapping)open source algorithm VINS-Fusion,develop a localization algorithm for mobile robots.In a scene where the visual feature points are missing,the problem of pose estimation failure will occur.In this paper,a detection algorithm for localization failure is added,and a constant speed model is proposed to assist the robot’s localization in the period of missing feature points.The problem of localization failure of the algorithm in this scenario is improved,and the robustness and applicability of the algorithm is improved.In a large outdoor scene environment,the robot can receive Beidou localization data by carrying the Beidou module.Combining the pose of the stereo+IMU and the Beidou localization data can obtain a smoother pose in the Beidou coordinate system to meet the localization and navigation requirements of the robot in the Beidou coordinate system.(5)Building a mobile robot to verify the localization algorithm by experiment.For different algorithms,the robot is equipped with different sensors to verify the precise localization algorithm at the target point,the AMCL localization algorithm fused with visual mask information,the improved stereo vision fusion IMU localization algorithm,and the VIO fusion Beidou localization algorithm.