Real-Time AGV Localization and Dense 3D Map Reconstruction Using Visual SLAM

Traditional Automated Guided Vehicle (AGV) navigation systems face challenges such as cumulative localization errors, high hardware costs, and reliance on pre-installed landmarks. To address these limitations, this study proposes a real-time AGV localization and navigation system based on visual Simultaneous Localization and Mapping (SLAM). The system integrates ORB-SLAM for robust pose estimation and REMODE for probabilistic dense depth estimation, enabling precise localization and high-resolution 3D point cloud map reconstruction. Implemented on the Robot Operating System (ROS), the framework ensures modularity and compatibility with mobile robotics applications. Validation was conducted through Gazebo-based simulations and physical experiments using and educational AGV platform. Results demonstrate real-time performance (30FPS for 720P images) and accurate dense 3D mapping in both static and dynamic indoor environments. The system's ability to operate with low-cost monocular cameras and its scalability for industrial automation, logistics, and healthcare applications highlight its practical significance. Future work will explore fusing with depth sensors to enhance robustness in complex scenarios.