Purpose Two-dimensional (2D) Lidar SLAM has been widely applied in indoor service robots. However, single-plane 2D LiDAR suffers from repetitive point-cloud patterns and scarce geometric features in corridor-like environments. These challenges lead to inaccurate loop closure detection and significant trajectory errors, thereby affecting mapping performance. The aim of this research is to develop a robust loop closure detection framework to enhance the overall accuracy of SLAM. Design/methodology/approach The algorithm uses a visual Bag-of-Words model for loop closure detection, dynamically adjusting word weights for corridor environments to enhance recognition. It incorporates Speeded-Up Robust Features-based geometric checks to eliminate false closures. The loop closure results are coupled with submap node poses for correction, and parameters are dynamically adjusted to improve trajectory tracking and mapping. Findings The proposed algorithm was validated and compared in three scenarios: a circular structure in the Gazebo simulation environment, a non-circular structure in the OpenLORIS data set and a real-world scenario. The results of this study demonstrate that the proposed algorithm achieves more stable mapping and lower trajectory errors in various long corridor environments compared to other algorithms. Research limitations/implications The algorithm performs better in identifying the closed loop of the circular corridor. It is more effective in reducing trajectory errors and enhancing mapping results. However, its recognition effect is relatively poor in non-circular corridor environments. Compared to the improvement in trajectory errors and mapping results in circular corridors, it is not obvious in non-circular corridor environments. In the future, lightweight semantic recognition or other methods can be considered to enhance the closed loop detection algorithm’s closed loop recognition effect in this type of non-circular corridor. Originality/value This study integrates a visual loop closure algorithm with traditional 2D LiDAR Simultaneous Localization and Mapping (SLAM) technology, optimizing it for structured environments such as corridors. By using a weighted visual bag-of-words model and Speeded-Up Robust Features-based geometric consistency checks, the accuracy of loop closure detection is enhanced. The detection results are used to correct the backend node poses and dynamically adjust parameters. This improvement enhances the robot’s mapping capabilities in corridor environments within acceptable limitations, providing a solution for more precise and stable SLAM applications in such structured scenes.
Tang et al. (Thu,) studied this question.