Research On Multi-robot SLAM Algorithm Based On Graph Optimization

Simultaneous Localization and Mapping(SLAM)is the basis for mobile robots to work autonomously in an unknown environment.Mobile robots with this technology can perform more complicated tasks such as navigation,path planning and exploration.In recent decades,researchers have proposed a variety of effective SLAM algorithms for single robot.However,when facing with a large-scale work environment or a task that requires high efficiency,the single robot SLAM cannot quickly and accurately create a global map.Therefore,the research on multi-robot SLAM has important significance in practical applications.Compared with single robot,multi-robot SLAM has the advantages of high work efficiency,high accuracy in creating maps,and high stability in complex environments.However,multi-robot SLAM also needs to solve the problems similar to single robot,such as relative pose acquisition and map fusion.Firstly,to solve these problems,this paper analyzed several commonly used environmental map description methods and studies two kinds of single robot SLAM algorithms.Secondly,a grid map fusion algorithm based on the maximum common subgraph is proposed to solve the grid map fusion without any prior information.According to the similarity between multiple sub-map sequences,map fusion problems construct the front-end of multi-robot SLAM,and establish a global optimization objective function to achieve back-end optimization;Finally,collected data by using the experimental platform,combined with the public data set,to verify the proposed algorithm.The main research contents of the thesis are as follows:(1)Analyzed and compared the advantages and disadvantages of grid map,feature map and topological map,and studied the filter-based SLAM algorithm and graph-optimized SLAM algorithm to compare the similarities and differences between the two algorithms theoretically.By analyzing the advantages and disadvantages of the two kind of algorithms,the graph SLAM is determined as the basis of multi-robot SLAM algorithm,and the environment is described by grid map.(2)Proposed a grid map fusion algorithm based on maximum common subgraph.Firstly,the ORB algorithm is used to extract the feature points and descriptors of the fusion map,and the feature points will be clustered.Secondly,the Hamming distance is used to calculate the initial matching of the feature points.Thirdly,through the spatial positional relationship of the feature points,the maximum common subgraph is searched by the virtue of backtracking method.Finally,the transformation matrix is calculated according to the feature point matching relationship in maximum common subgraph,and the grid map fusion is implemented according to the grid map fusion method.(3)Analyzed and compared the multi-robot architecture,combined with the actual computing power of the subject robot,and use the distributed method to study the multi-robot SLAM.Based on the graph optimization method,the multi-robot SLAM research is divided into front-end and back-end.Based on the grid map fusion algorithm of this subject,a multi-robot relative pose acquisition and closed-loop detection method are proposed.Based on the constraint information provided by the front-end,combined with the bundle adjustment,the multi-robot SLAM optimization objective function is constructed,and the sub-map matching relationship is optimized to create a global map.(4)Set up an experimental platform to move the Rplidar A1 laser radar to the Turtlebot2 mobile platform.Studied the ROS framework and use the communication between the topics to save the radar information and the car sensor information to create a sub-map sequence.According to the created sub-map sequence and public data set,the validity of the multi-robot SLAM algorithm proposed in this paper will be tested.