Research On The Method Of Map Construction With Multi-Robots Based On Graph Optimization

In recent years,the SLAM algorithm for a single robot in a small scene has been gradually developed and applied.Due to the limitation of the computing power,the low efficiency of localizing and mappping in a large scene forms the bottleneck.Multi-robot collaborative mapping is an effective way to solve this problem,but current methods generally face problems like low information interaction efficiency,low map accuracy,and mismatching in the map fusion process.To this end,this paper proposes a method to construct the map by multi-robot based on graph optimization under the 'cloud-edge-end'network architecture.The specific research content and innovative work are as follows:(1)Aiming at the complexity and low efficiency of information interaction between multi-agents,the multi-terminal communication mode under the 'cloud-edge-end'architecture is designed and implemented,and the connections between robots,the edge server and the cloud platform were established.Information subscription is completed by LZ4 algorithm,and the data interaction process is simplified through the publishing and sucribing of information,which realizes the effective distribution of computing tasks and the distributed exploration and centralized processing of multiple robots system.The experimental results show that the communication delay across network segments can meet the real-time data transmission requirements of multi-robot collaborative mapping,and the bandwidth occupancy is reduced by 30%,and the computing resource consumption of a single robot is reduced by about 20%on average.(2)Aiming at the problem of low accuracy of multi-robot collaborative mapping,the algorithm of sub-map construction and global map optimization are improved.During the construction of each sub-map,the front-end odometry is generated by the method of Normal Distribution Transformation,and characteristic landmarks are used to find the accurate loop closure in different environments.Finally the factor graph model is constructed and optimized by fusing multi-source sensor information.Based on the fusion of each sub-map,the initial global map is generated and continuously updated with the incremental data from each robot.Furthermore,the graph model fusion is used to obtain the global factor map,and the global map is further optimized and adjusted through back-end optimization.Experiments in real scenes have shown that the map accuracy has increased by an average of 5%,and the map correction is effecitve for scenes which are difficult for other algorithms to build the map.(3)Aiming at the mismatching during the fusion of the global map,a method based on environmental feature landmarks is proposed for map merging.During the process of sub-map construction,landmark detecting and coordinate binding are performed to simplify the task of feature matching.With the unique landmark representations used as reference points to realize the association of multiple sub-maps.In addition,the landmark coordinates are used to determine the transformation relationship between multiple map coordinate systems and obtain the optimal solution to achieve map alignment and data fusion.The test results show that the robot can realize the rapid association of maps in multiple scenes and improve the matching accuracy of global map fusion.