Automating on-site object inspection with a quadruped robot and BIM

Abstract

Deploying robots into construction sites and built environments using Building Information Modeling (BIM) is difficult due to environmental changes and inconsistencies between BIM and real-world structures. To overcome these challenges, an integrated framework is proposed for automated inspection featuring adaptive BIM-driven navigation for a quadruped robot. The framework addresses challenges of localization, mapping, and navigation in challenging environments. By leveraging BIM for initialization and planning, it dynamically verifies BIM elements through sensor data, updating the map to distinguish between fixed structures, missing elements, and dynamic obstacles. This enables robust navigation despite environmental uncertainties. The capability supports automated tasks across the building lifecycle, demonstrated through object inspection using adaptive next-best-view (NBV) planning and vision–language model (VLM) analysis without task-specific training. Both simulated experiments and real-world validation confirm strong performance despite dynamic obstacles and BIM discrepancies. The approach enhances robotic adaptability in construction and facility management contexts while reducing manual inspection requirements. • BIM-initialized 3D LiDAR SLAM enables robot localization with respect to BIM. • 3D mapping detects BIM-defined structures and obstacles for adaptive navigation. • Next-best-view planning adapts to occlusions and BIM/reality layout differences. • Tested in simulation and validated in the real world with Boston Dynamics Spot robot. • Successful operation despite BIM discrepancies and dynamic obstacles in testing.