Research On 2D LiDAR SLAM High Precision Mapping Based On Control Network Constraints

Location Based Services(LBS)has evolved into an important part of the modern service industry,and high-precision maps are key to supporting the rapid development of the location services industry.How to develop a large-scale indoor mapping technology with high precision,low cost and fast collection of geographic information data has attracted more and more attention.SLAM(Simultaneous localization and mapping)is a hot topic in the field of robotics.SLAM address the problem of acquiring a continuous map of a mobile robot environment based on the environment-aware sensor while simultaneously localizing the robot relative to the map.SLAM unifies positioning and mapping under a framework,which is an effective way to solve the problem of mobile mapping.After years of development,the SLAM technology has been relatively mature,and the filter-based SLAM method has been gradually replaced by the SLAM method based on graph optimization.The SLAM method based on graph optimization can achieve better results in the case of adding loop closure constraints or in relatively smallscale environments.However,on the one hand,if there are no loop closure constraints,the positioning and mapping accuracy is difficult to improve because the cumulative error of SLAM is difficult to eliminate.On the other hand,with the expansion of the mapping area,the SLAM faces many difficulties,such as loop closure detection,the large amount of calculation,the large memory occupation,and the limited mapping accuracy.In order to solve these two problems,this paper proposes the corresponding solution and verifies the effectiveness of the proposed solution through experiments after expounding the basic theory of SLAM method based on graph optimization.The main work and contributions of this paper are:(1)For SLAM mapping scenes without loop closure conditions,such as the typical "U"-type mapping area,this paper proposes to add distance constraints to the back-end of graph SLAM to eliminate error accumulation and improve mapping accuracy.The experimental results show that the SLAM error accumulation can be effectively eliminated only if the distance constraints form a stable geometry such as a triangle.Therefore,we connect the control points according to the Delaunay triangulation and measure the side lengths as the distance constraints,the results show that the SLAM mapping result of adding Delaunay triangulation constraints has similar performance with the SLAM mapping of adding all distance constraints formed by interconnecting control points.In the experiment,the accuracy of the SLAM mapping result for a typical mapping area without loop closure conditions("U"-type region,the path length is more than 300 meters)was increased from 1.65 m to 0.36 m,and the accuracy was significantly improved by about 79%.(2)For the SLAM mapping problem in large scale environment,distributed SLAM adopts a divide and conquer method.Different from previous distributed SLAM methods,this paper implement distributed SLAM by adding control network constraint information to the back-end of graph SLAM optimization for the first time,more importantly,the SLAM mapping results for each partial map are in the absolute coordinate system.The results indicate that adding control network constraints can significantly improve the accuracy of SLAM mapping results,both in absolute accuracy and relative accuracy.The field test in a large scale parking lot shows that the relative accuracy can achieve 0.08 m,which is improved by 49.8% compared to the SLAM result without any constraints,and the absolute accuracy can achieve 0.126 m.The field test shows that adding control network constraints to the back-end of graph SLAM can achieve almost the same accuracy of SLAM mapping result with loop closure constraints,it indicates that adding control network constraints to the back-end of graph SLAM can solve the SLAM problem without loop closure conditions.In the experiment,we found that the number of control points has a certain influence on the accuracy of SLAM mapping results.When the number of control points reaches a certain level,adding more control points will not significantly improve the accuracy of SLAM mapping results.Therefore,it is necessary to select a reasonable number of control points and a control network structure according to actual conditions in practical applications.In summary,for the SLAM mapping problem without loop conditions and the SLAM mapping problem for large-scale environment,this paper proposes the corresponding solution and verifies the effectiveness of the method through experiments.