Enhancing mapping precision in autonomous delivery robots through tightly-coupled fusion of uncertainty-aware GPS and LiDAR odometry

Empowering delivery robots with enhanced mapping is crucial for ensuring seamless autonomy in autonomous navigation. Achieving precise position estimation requires a comprehensive three-dimensional point cloud map of the sidewalk environment. However, in the process of mapping extensive urban regions using current techniques, the occurrence of mapping trajectory misalignment can result. It causes gradual inaccuracies in the map that accumulate over time, raising concerns about relying on distorted or improperly aligned maps. The paper presents an approach that merges GPS data to provide improvement in 3D mapping, thereby mitigating the accumulation of errors. The outcomes of our proposed technique exhibit exceptional performance in both quantitative and qualitative assessments when compared with existing methods. The proposed approach also introduces a machine learning-based system that is sensitive to uncertainty, allowing it to switch to the LIO system in situations where GPS data is less reliable. Lastly, the paper introduces an innovative approach to scanning through and filtering point clouds linked with dynamic objects on the map. In summary, the primary objective of this paper is to create an extremely accurate map for further localization and navigation.