Research On Collaborative Mapping Technology Of Air-ground Robots Based On Multi-sensor Fusion

In recent years,the intelligence level of mobile robots has been greatly improved,and they have played an important role in production and life,modern operations,disaster rescue and other scenarios.The navigation system of a mobile robot includes three modules: autonomous positioning,environment perception,and planning control.The environment map is the basis for the realization of these three modules.Therefore,the establishment of a map of the surrounding environment is crucial for the operation of the mobile robot.Aerial robots can quickly obtain a global top-down map by taking advantage of the viewing angle,but it is easy to form blind spots in the surrounding area of buildings,so ground robots can be used to make up for its deficiency.Aiming at the limited field of view of a single type of robot and the limited perception ability of a single sensor,this thesis proposes a collaborative mapping method for air-ground robots based on multi-sensor fusion to improve the robot’s mapping ability.The main research contents are as follows:(1)A multi-level map construction method for aerial robots is proposed.Aerial robots use lidar and IMU for data collection,and are used for the construction of point cloud maps,octree maps,and grid maps.Through the multi-DOF data collection of the gimbal,the flight time and flight distance of the aerial robot are reduced,and the constructed multi-level map can be used for the 2D and 3D navigation scenarios of the ground robot.In order to construct incremental dynamic maps,an adaptive random sampling consensus algorithm is used to automatically correct the tilt problem of maps constructed by FAST Lidar-Inertial Odometry.(2)Research on automatic navigation and mapping of ground robot.After the ground robot obtains the grid map established by the aerial robot,it uses the Dijkstra algorithm of the navigation stack and the dynamic window method for path planning.In the process of automatic navigation,it uses the costmap for real-time obstacle avoidance,and uses the FAST Lidar-Inertial Odometry algorithm to build the map.While improving the automation of the mapping process of the ground robot,it also avoids the possible on-site hazards when manually collecting data.(3)A map data fusion algorithm based on point cloud registration is proposed.Firstly,the map data is preprocessed by filtering algorithm to obtain the source point cloud and target point cloud for registration;then the features of the two are obtained through HARRIS algorithm and Signature of Histogram of Orientation(SHOT)descriptor;Finally,the random sampling consensus algorithm is used to obtain the coarse registration results and the Iterative Closest Point(ICP)algorithm is used to obtain the fine registration results.The registration method can be used for map data fusion between open-air and ground robots,so as to realize collaborative map construction.This thesis makes use of the respective advantages of air-ground robots for collaborative mapping,and constructs a complete environment map efficiently and accurately,which makes up for the insufficiency of single robot mapping and improves the automation of the mapping process.This thesis contains 64 figures,17 tables and 83 references.