Enhancing Automated Guided Vehicle Navigation with Multi-Sensor Fusion and Algorithmic Optimization

Automated Guided Vehicles (AGVs), as a novel type of mobile robot, have significantly enhanced the operational efficiency of mobile robots in industrial environments. The navigation and obstacle avoidance of AGVs based on 2D-Lidar lack the capability to detect the category of obstacles. To address this, our paper deploys a Yolo object detection algorithm based on MobileNet for the detection of pedestri-ans and obstacles in specific tasks, thereby improving obstacle avoidance capabilities. Furthermore, many SLAM algorithms prioritize mapping accuracy and error reduction, neglecting the importance of relocalization speed in industrial applications. This paper introduces an improved relocalization strategy based on the Cartographer SLAM algorithm and employs a multi-level dimensional extraction approach for enhancing the precision of GridMap in pure vision-based SLAM 3D point clouds. Ultimately, with the incorporation of a dual-laser Lidar extrinsic calibration algorithm, the relocalization time of the enhanced Cartographer algorithm is reduced by 35%. This improvement in obstacle avoidance and object detection capabilities ensures the safety and stability of AGVs in snecialized scenarios.