Research On Multi-sensor Fusion SLAM Algorithm Based On Loose And Tight Coupling

In recent years,with the widespread use of robots in industrial production and daily life,robot-oriented SLAM(simultaneous localization and mapping)technology has become one of the focuses of academic research.When the robot runs in an unfamiliar environment for a short time,most of the current SLAM algorithms can more accurately estimate its own pose and environment map.Due to the noise in the observation data of various sensors,as the system running time increases,the robot trajectory will drift,and loop detection and relocation optimization are required to eliminate the cumulative error in the pose.At present,the use of visual sensors in SLAM is relatively mature,but SLAM relocation that only uses visual information will be unstable in many scenarios,resulting in a decrease in positioning and map construction accuracy.As a result,it is a current development trend to introduce multi-sensor data to complement and constrain to achieve high-precision and high-reliability SLAM system positioning.In this paper,by introducing GPS global information,a multi-sensor tightly coupled and loosely coupled SLAM framework with dynamic feedback mechanism,TL-SLAM,is proposed to realize high-precision UAV trajectory estimation and map construction.(1)In the local map,we construct the reprojection residual model,the IMU incremental residual model and the global position residual model respectively according to the observations obtained by the vision,IMU and GPS sensors,and perform joint nonlinear tightly coupled optimization for the pose of the robot.(2)In order to further eliminate the cumulative error in the local pose,this paper uses the loosely coupled model to fuse the local map information and GPS observations in the global map,constructs the absolute residual model and the pose relative transformation residual model,and uses the least square multiplying constraints to minimize the residuals to obtain higher-precision global pose information when relocation is not possible.Algorithm verification is carried out on multiple difficulty level data sequences of public data set Eu Ro C.And conducted trajectory comparison experiments with the other state-ofthe-art SLAM frameworks(including VINS-MONO,ORB-SLAM2,ROVIO,OKVIS,MSCKF),and quantitatively analyze the accuracy of pose estimation.In addition,verification experiments are conducted on the large outdoor driving dataset KITTI.The results show that the trajectory of the proposed algorithm is closer to the real value,and its pose accuracy is better than all comparison algorithms,and it has strong competitiveness compared with the existing open source framework.