Hierarchical Factor Graph-based Localization and Mapping leveraging Building Information Models

The construction industry is undergoing significant digital transformation through the convergence of Building Information Modeling (BIM) and robotic automation. However, the deployment of autonomous robots in construction environments remains challenging due to the complexity of these dynamic spaces and the limitations of traditional geometric SLAM approaches. This thesis addresses the fundamental challenge of developing robust hierarchical factor graph-based mapping and localization systems for mobile robots that effectively leverage BIM knowledge while handling real-world deviations between “as-planned” and “as-built” environments. This research introduces a novel framework comprising three interconnected algorithmic contributions: Situational Graphs (S-Graphs), informed Situational Graphs (iS-Graphs), and deviations informed Situational Graphs (diS-Graphs). S-Graphs establishes a hierarchical 3D scene graph representation that organizes construction environments into semantically meaningful layers, including walls, rooms, corridors, and floors, achieving an average mapping accuracy of 20.9 cm RMSE in real-world construction sites. The framework transitions from basic geometric understanding to structured semantic representation, enabling direct comparison against BIM models for "as-built" documentation. Building upon this foundation, iS-Graphs integrates architectural prior knowledge through a novel graph matching approach that enables global localization without prior mapping. By representing both BIM models and robot perception as hierarchical factor graphs, the system achieves explicit one-to-one correspondences between structural elements, demonstrating superior performance with point cloud RMSE values between 17 cm and 21 cm in real-world construction sites while outperforming traditional methods. The culminating contribution, diS-Graphs, addresses the critical real-world limitation of plan-reality deviations through simultaneous localization, mapping, and deviation detection. This algorithm performs three
interconnected tasks: tightly coupling SLAM factor graphs with architectural plans, achieving global localization despite plan inaccuracies, and detecting structural deviations in real-time. diS-Graphs achieves a 43% improvement in localization accuracy over existing methods while successfully detecting translational deviations up to 35 cm and rotational deviations upto 15 degrees , covering the most realistic construction tolerances. This research makes significant contributions to the field of localization and mapping for mobile robots, especially in construction environments. It connects theoretical academic models with practical, deployable technology, laying the groundwork for the future of intelligent, spatially-aware robotic systems in construction environments. The thesis demonstrates that hierarchical factor graph-based approaches combined with BIM integration can significantly advance robotic situational awareness in construction environments, with implications extending beyond robotics to construction quality control, building inspection, and intelligent facility management.