Large-Scale Hybrid Metric-Topological Map Building And Maintenance For Mobile Robots

Long-term autonomous operation is the basis for mobile robots to work in outdoor dynamic environments.In real-world applications,the environment around the robot will change with the seasons and human factors,and the reliability of the prior map will be reduced when there is a great difference between the historical scene perception data in the prior map and current scene perception data.In this paper,we studied and proposed solutions to the problem of hybrid metric-topological map building and maintenance for mobile robots in outdoor dynamic environments.To achieve simultaneous pose estimation and map building with high precision in large-scale outdoor environments,we propose an integrated GNSS/Li DAR-SLAM pose estimation framework,to overcome the problem of intermittent loss of GNSS signals.While working in open scenes,we propose an auto coordinate alignment algorithm,which aligns the coordinate system between Li DAR and GNSS online without relying on other equipment.Then,we use Kalman Filter to fuse GNSS data and IMU data.The fused data is used as initial values for the back-end optimization algorithm based on graph optimization to perform the real-time pose estimation.While working in GNSS-denied scenes,we propose a backward-adjustment method and graph optimization algorithm to eliminate the drift error of pose estimation.To improve the efficiency of long-term large-scale map maintenance of mobile robots in high dynamic environments,a hybrid metric-topological map model is used to describe the environment.To improve the accuracy of the local metric map at the topological node,we propose a clustering algorithm with dynamic threshold adjustment to cluster 3D point clouds and removes the outlier points based on clustering results.Dynamic objects are detected based on the contrast between two continuous frames of data.To improve the accuracy of the topology to describe the environment,the degree of difference in different scenes,whether the robot is located at the road intersections,the distances from the previous node are fused to build cost function to judge whether the position of the robot can be set as a topological node.For the problem of long-term maintenance of the hybrid metric-topological map,we use a global descriptor for relocalization.A new topological node will be added to the hybrid metric-topological map when relocalization fails.We remove the redundant topological node which is detected based on the relocalization success rate and the length of time from the construction to avoid unlimited growth in the number of topological nodes.In this paper,we use a self-developed mobile robot platform to demonstrate that the integrated GNSS/Li DAR-SLAM pose estimation framework can perform real-time pose estimation with high accuracy in partially GNSS-denied outdoor environments.We also tested the hybrid metric-topological map building and maintenance algorithm based on the NCLT public data set and the self-collected data set.The experimental results show that our method is valid,and the maintenance of the hybrid metric-topological map can effectively improve the relocalization success rate.In addition,deleting redundant topological nodes can effectively reduce the number of topological nodes while ensuring the success rate of relocalization.