Research On Air-ground Multi-robots Simultaneous Localization And Mapping Algorithm Integrating Visual Features

SLAM aiming to solve the problem of mobile robot simultaneous localization and mapping in unknown environment,is the foundation of robot subsequent navigation planning,automatic driving,and various complex tasks.In recent years,both laser SLAM and visual SLAM applied on single robot have gradually become matured.However,when dealing with large-scale indoor environment,single robot will encounter problems such as low efficiency and locating drifting,which makes it difficult to perform fast and accurate map construction.Unlike single robots,which have weak sensitivity in complex environment,air-ground multi-robots SLAM can have a wider coverage of environment and better ability to explore the environment,match more diversified data,and better realize the robot function complementary.This paper will study on feature extraction,data association and map merging of air-ground multi-robots SLAM.Firstly,this paper introduces the main SLAM methods for single robot and analyzes the advantages and disadvantages of topological map,point cloud map and raster map.Also,Octomap is selected to describe the environment.By comparing several classic laser SLAM and visual SLAM methods,including GMapping,Loam and Cartographer,LSD SLAM and RGBD SLAM.We finally select Cartographer(based on scan-to-submap)and RGBD SLAM(based on ORB features)as the research foundation of single robot SLAM for UGV and UAV respectively.Secondly,this paper proposed two methods:the corner extraction method of grid map based on sliding window and the map merging method based on maximum common subgraph.The former method builds a sliding window around the occupied grid center and constructs an evaluation function to solve the corner problem,calculating three-dimensional convolution values and using Principal Component Analysis(PCA)to determine the parameter interval.Thus,the feature of map can be extracted.As for the later method,Delaunay triangulation was used to divide the feature points and construct the initial match so that the optimal solution of the maximum common subgraph could be found.Therefore,grid map between robots can be merged according to the best transformation matrix calculated by singular value decomposition.Thirdly,we analyze the limitations of the map merging method based on MCS for airground multi-robots.According to the characteristics of the submaps created by UGV and UAV,a mathematical model of air-ground multi-robots submap integrating visual features is established.Thus,multiple constraints are constructed based on the submap model.By using the optimization method based on LM method,data association between UGV and UAV’s submap would be achievable.Finally,the global map for air-ground multi-robots is constructable.Finally,UGV and UAV are used to construct experimental platforms.Experiments are carried out in multiple experimental scenarios and open data sets are used to verify the effectiveness of the air-ground multi-robots SLAM method which is proposed in this paper.