This paper introduces VLW-Fusion, a tightly coupled SLAM framework that combines stereo vision, 2D LiDAR, and wheel odometry for robust localization and 3D mapping in mobile robots. To overcome the limitations of single-sensor SLAM, such as sensitivity to illumination, texture scarcity, and environment-specific errors, the proposed method employs a unified factor graph to fuse multi-sensor data. A sliding window optimization strategy is used in the front-end for computational efficiency, while pose graph optimization refines the global trajectory in the back-end. Additionally, a vertically mounted 2D LiDAR is synchronized with optimized odometry to progressively construct a 3D point cloud map. The proposed framework has been evaluated through various indoor and outdoor environments using a real robot platform. Experimental results show that VLW-Fusion achieves higher accuracy and mapping performance compared to single- and dual-sensor SLAM methods.
Hwang et al. (Mon,) studied this question.