Construction Robot Localization System Based on Multi-sensor Fusion and 3D Construction Drawings

This paper presents a localization system which uses multi-sensor fusion and 3D construction drawings to conduct high-precision localization in construction sites. The localization system is composed of global localization and position tracking. In the global localization stage, we propose an image-driven point cloud segmentation pipeline to filter the point cloud, and obtain the initial pose using a point cloud registration algorithm between the filtered point cloud and the 3D building model. In the position tracking stage, we propose to use unscented Kalman filter to fuse the point cloud registration pose with IMU, which helps to improve the accuracy of localization. This system is tested in realistic construction sites. We verify the feasibility of the proposed global localization algorithm and show that the system achieves higher accuracy than LiDAR SLAM localization.