Robot Trajectory Optimization and Online Semantic 3D Perception for Mobile LiDAR Scanning in GPS-Degraded Built Environments

ABSTRACT Integrating robots into the architecture, engineering, and construction industry is gaining traction, as robots facilitate mobile scanning, as-built 3D reconstruction, and digital documentation for facility management. However, deploying robots in built environments remains challenging because many indoor and outdoor spaces are GPS-degraded, and robot navigation is frequently affected by humans, temporary objects, and movable obstructions which disrupt mobile LiDAR mapping. This paper presents a unified method integrating robot trajectory optimization, online semantic 3D perception, and SLAM-based mobile LiDAR mapping for GPS-degraded environments including indoor, outdoor, and building-street interfaces. The proposed method formulates robotic scanning as a constrained 3D coverage and navigation optimization problem. Geometric 3D maps are used to optimize viewpoints and navigation trajectories by considering structural geometry, LiDAR field-of-view, surface coverage, path-intersection severity and robot operational constraints. During execution, semantic 3D perception is added as a dynamic constraint for local replanning. Humans are semantically segmented from LiDAR data and incorporated into a semantic cost map, enabling the local planner to update the robot trajectory with human-aware safety constraints. LiDAR scans along both the planned and adjusted trajectories are incrementally registered using SLAM to reconstruct point cloud 3D maps. Experiments indicate that the proposed method improves the spatial coverage, mapping efficiency, and point cloud data collection quality.