LiDAR localization using position-encoded landmarks without point cloud maps

LiDAR-based localization plays a critical role in autonomous driving and robotic navigation. However, traditional methods rely heavily on constructing high-precision point cloud maps, which is both time-consuming and labor-intensive. To address this, we propose an innovative localization approach that eliminates the need for point cloud maps by leveraging LiDAR and position-encoded landmarks. Our method encodes positional information into the shape of specially designed landmarks, strategically deployed in the environment. Subsequently, we fully leverage the advantages of LiDAR in accurately measuring distances and capturing the spatial structures of objects to detect and recognize the landmarks in the environment. By decoding the positional information embedded in the landmarks, precise vehicle localization is achieved. To overcome the limited information capacity of individual landmarks due to LiDAR’s reduced accuracy at long distances, we integrate multiple landmarks in a collaborative manner. By combining their encoded information and spatial relationships, we achieve high-precision localization without relying on point cloud maps. Experiments in CARLA and Autoware.AI simulators validate the effectiveness of our approach, offering a novel solution for LiDAR-based localization.