Improved Localization Algorithm Based on MultiSensor Fusion for Shopping Robots

Abstract

To address the issues of indoor shopping robots being unable to locate their initial position when starting from arbitrary locations and map mismatches during navigation, an improved ORB-SLAM3 positioning system that fuses LiDAR technology with RGB-D has been designed. First, LiDAR provides limited positioning information, whereas depth cameras offer a broader range of positional data. Therefore, a system that fuses LiDAR with RGB-D cameras is proposed, introducing visual information to merge visual feature maps with two-dimensional grid maps, enabling mode switching between LiDAR and visual sensor positioning. Secondly, the 6-DoF in the visual algorithm is simplified to 3-DoF, effectively reducing positioning time. Finally, through positioning experiments, it has been validated that the positioning accuracy with mode switching improves by 58.5% compared to using LiDAR alone, the positioning time accuracy improves by 35.13% compared to the ORB-SLAM3 algorithm, and the positioning error accuracy improves by $\mathbf{2 6. 4 2 \%}$ compared to the ORB-SLAM3 algorithm, demonstrating the effectiveness of the improved positioning algorithm.