Mapping, Navigation, Dynamic Collision Avoidance and Tracking with LiDAR and Vision Fusion for AGV Systems

The Automated guided vehicle (AGV) has been one of the most popular topics for the last few decades, on account of the industrial need for higher efficiencies. Four basic functions are required for an AGV system: mapping, navigation, dynamic collision avoidance, and coordination. The most commonly applied system is the 2D Simultaneous Localization and Mapping (2D SLAM) system which utilizes a 2D LiDAR at a relatively low cost and high efficiency. However, the 2D SLAM algorithm has a critical defect. It can only acquire 2D information, leading to some obstacles being ignored. This article aims to apply a LiDAR and vision fusion algorithm with an RGB-D camera. The specific data fusion algorithm selected is RTAB-MAP. Key issues encountered in the general implementation of the algorithm are tackled with comprehensive experiments. The original 2D SLAM and the fusion SLAM algorithms are tested and compared to reveal ways to further improve the system design. By using data fusion SLAM system, the mapping efficiency of the AGV in three specific scenarios is improved including the environment with low obstacles, thin obstacles, and hanging obstacles.