Research On SLAM Technology Of Mobile Robot

Since the advent of slam technology in 1986,after 30 years,from the initial "probability era" to the "graph optimization era" and now to the "robustness and prediction era",algorithm theory and hardware performance have been greatly improved,and slam technology has gradually realized the technical landing.At present,slam technology is mainly divided into vision slam technology and laser slam technology.The main sensor of vision slam technology is the camera,because the light,sharp shaking and less texture will lead to the loss of camera motion tracking.Therefore,the slam technology based on single vision is still in the research,and can not be used commercially.However,The technology based on laser slam,because of the stability of lidar ranging and the maturity of technical theory,the development and application based on single laser slam technology has gradually entered human life.With the needs of mining operations,industrial operations and underwater exploration,slam technology relying on a single sensor is difficult to meet the positioning in complex environment.In order to make up for this defect,slam technology based on multi-sensor fusion has become the mainstream method to solve the problem.Through the information fusion of multiple sensors,the strong sensing ability of agent in complex environment is guaranteed.Based on the ROS platform,the multi-sensor laser slam technology is studied,and a calibration method of dual 3D lidar and a loosely coupled multi-sensor map optimization framework are proposed.The main research contents are as follows:(1)In the traditional laser calibration,it is difficult to find the constraint relationship between laser points,and the calibration process needs additional sensors and artificial markers for assistance.This paper studies the calibration of vehicle borne dual 3D lidar.The corner shaped scene is selected as the calibration target,and the three linear independent planes are matched by the least square method,Based on Kabsch algorithm,the rotation matrix and displacement matrix of the dual lidar system are obtained.The Levenberg Marquardt(LM)algorithm is used to iteratively optimize the rotation matrix and displacement matrix,and the calibration results are obtained.(2)The front-end odometer based on laser point cloud feature descriptor matching is studied.Slam technology needs real-time,and it is impossible to match all the point clouds.By calculating the curvature of the point cloud,the representative features:plane features and corner features are selected for matching,and the odometer estimation is obtained.Then the two-step LM algorithm is used to reduce the error of odometer estimation.(3)Based on the work of(2),a loosely coupled multi-sensor graph optimization framework is studied.In(2),the odometer motion is optimized by nonlinear iteration.With the long movement time,only lidar information is not enough to eliminate the accumulated error,so it is necessary to provide auxiliary information through other sensors to construct the odometer motion pose map,To optimize the back-end,this topic optimizes the odometer movement through the rotation pose constraint provided by IMU and the displacement vector constraint of GPS.(4)This paper studies the loopback detection based on kdtree search.By setting the search radius,the loopback detection within the radius can be searched with the current frame construction history constraints,and the global consistent optimization of the map and pose being constructed is carried out to obtain the optimal pose map and environment map.