Research On Multi-Robot Cooperative SLAM Method Based On Multi-Sensor Fusion

In recent years,mobile robots have been increasingly used in industrial,agricultural,and service industries,and Simultaneous Localization and Mapping(SLAM)is one of the key technologies for mobile robots to achieve autonomy,thus attracting much attention.As the application scenarios of mobile robots become more diverse and complex,the challenges for individual robots to perform SLAM tasks are increasing,especially in large-scale scenarios or work situations with high efficiency requirements.Multi-robot collaborative SLAM is an effective solution for such scenarios and situations.In this article,we purposefully analyze and study the key points and challenges of multi-robot SLAM,and implement a multi-robot SLAM approach for outdoor scenarios.The main contents and contributions are as follows:Firstly,the single-robot SLAM algorithm is studied.The SLAM technology with LiDAR-Only cannot solve the problems of large robot motion changes and degradation.This article adopts the preintegration value of Inertial Measurement Unit(IMU)as the constraint to correct the pose estimation,and the GPS measurement as the loop closure judgment.Then,the laser odometry,keyframe filtering,and loop closure modules are established.A factor graph optimization is employed for precise real-time pose estimation and map construction for a single robot.The method is verified on public datasets.Secondly,the problem of multi-robot pose optimization and map merging in a distributed framework is studied.In this paper,a loop detection method combining laser point cloud descriptors and GPS data is proposed to address this problem.In order to effectively deal with the inevitable mismatching in the loop closure detection stage,a pairwise consistency maximization algorithm is applied to further filter out the loop closures that satisfy each other’s consistency and achieve the purpose of eliminating mismatching.For the optimization of multi-robot pose,the established loop closures and robot-related pose are used to construct a factor graph,and the pose is continuously adjusted through optimization to align the local maps and generate a global map.Finally,a mobile robot platform is built to collect datasets for the multirobot SLAM task in the campus.The proposed method is analyzed and demonstrated in terms of robot trajectory estimation accuracy and map merging efficiency for localization and map building in large-scale outdoor scenarios.