A Hybrid LiDAR-Based Mapping Framework for Efficient Path Planning of AGVs in a Massive Indoor Environment

Path planning is a computation-consuming task for autonomous ground vehicles (AGVs) in a massive indoor environment, especially for AGVs having limited computing resources. Popular path planning methods based on metric maps, such as occupancy grid maps, are computationally expensive in such environments. Therefore, this article proposes a hybrid mapping framework integrating the state-of-the-art metric mapping method and the devised topological mapping method, whereby AGVs can realize efficient path planning. The key idea is to select critical landmarks (such as turns, doors, and junctions) as topological vertices and build a succinct topological map superposing on the metric map. To realize this, a simple and interpretable feature for landmark recognition is proposed. Experiments conducted in simulation, datasets, and real-world environments have proven the framework's ability to perform hybrid mapping in massive environments with areas of up to 30 000 m2. Furthermore, compared with other typical hybrid mapping methods, the proposed framework considers critical landmarks and surpasses them in the hybrid map's succinctness. With this framework, in contrast to metric maps, the search space for path planning is reduced by orders of magnitude. As a result, path planning on the hybrid map is at least two orders of magnitude faster than representative methods, meanwhile realizing comparable performance with them.