Research On Localization Algorithm Based On Multi-sensor Fusion For Indoor Robots

After years of development,robots have a variety of functions,among which positioning and navigation function is the premise of mobile robots to achieve other functions,which has great research value and practical significance.Most indoor mobile robots rely on autonomous positioning to achieve their functions.The autonomous positioning technology of robots has achieved considerable development,but it is still vulnerable to the influence of the surrounding environment and cannot achieve the positioning function completely independently and accurately.The autonomous positioning schemes based on different sensors have different error sources and application scenarios.In order to achieve stable positioning performance of mobile robots in indoor planar scenes,this project integrates sensor information such as single-line lidar,depth camera and inertial measurement unit for self-positioning of mobile robots,which improves positioning accuracy and robustness of robots in different indoor scenes.The specific research contents are as follows:The motion models of robot,wheel odometer and inertial measurement unit were established,and the errors were analyzed.The data of odometer and inertial measurement unit were initially fused by diffusion Kalman filter algorithm,which provides reasonable initial values for laser and visual positioning system.The existing laser and visual positioning principles were analyzed and their advantages and disadvantages were verified by experiments.Through data fusion,the problems of mismatching in laser loopback detection and the sensitivity of visual positioning system to environment were solved.The parallel front end and coupling back end of laser and visual positioning technology were adopted to improve the response speed,positioning accuracy and positioning performance of the system in degraded environment.Based on the graph optimization theory,the nonlinear pose optimization was carried out by using visual feature constraints and laser feature constraints,which avoided the problem that traditional filtering methods could not optimize the past pose.The loopback detection link of laser and visual positioning system was run in series,which makes full use of the characteristic information of word bag model and the advantages of laser computation,and reduces the false matching rate at the loss of certain matching success rate.Finally,the proposed positioning method was verified on the McNamm wheel robot experimental platform,and the positioning accuracy,relocation accuracy and adaptability of the Cartographer algorithm and the proposed multi-sensor fusion localization algorithm were tested and compared in indoor closed and open scenarios,respectively.The experimental results showed that the laser vision fusion method has higher positioning accuracy than the pure laser or visual positioning method in different scenes,and could effectively distinguish similar scenes,and achieved loop detection and repositioning stably.