BIM-SLAM: Integrating BIM Models in Multi-session SLAM for Lifelong Mapping using 3D LiDAR

TL;DR

The paper tackles the challenge of lifelong indoor mapping in GPS-denied environments by leveraging available BIM models as a global reference. It introduces BIM-SLAM, a modular three-step pipeline that (i) generates BIM-derived session data, (ii) performs ground-truth multi-session anchoring to align new data with the BIM model, and (iii) constructs an aligned map with change detection to identify new elements. The approach enables accurate BIM-aligned maps without requiring a known initial pose and supports visualization of both existing BIM elements and newly observed objects through voxelized surface reconstructions. Experimental results in simulation and the real world demonstrate improved alignment to BIM and effective change detection, highlighting the method’s potential for long-term map management and situational awareness in complex buildings.

Abstract

While 3D LiDAR sensor technology is becoming more advanced and cheaper every day, the growth of digitalization in the AEC industry contributes to the fact that 3D building information models (BIM models) are now available for a large part of the built environment. These two facts open the question of how 3D models can support 3D LiDAR long-term SLAM in indoor, GPS-denied environments. This paper proposes a methodology that leverages BIM models to create an updated map of indoor environments with sequential LiDAR measurements. Session data (pose graph-based map and descriptors) are initially generated from BIM models. Then, real-world data is aligned with the session data from the model using multi-session anchoring while minimizing the drift on the real-world data. Finally, the new elements not present in the BIM model are identified, grouped, and reconstructed in a surface representation, allowing a better visualization next to the BIM model. The framework enables the creation of a coherent map aligned with the BIM model that does not require prior knowledge of the initial pose of the robot, and it does not need to be inside the map.

Figures (6)

• Figure 1: Overview of BIM-SLAM. The pipeline consists of three steps: BIM-based session data generation, Multi-session anchoring, and Map update.
• Figure 2: Generated SD from the BIM model. In the top left, the ground truth pose graph-based map $\mathcal{G}_{GT}$, with its nodes and edges (the result of odometry constraints). Besides having position and orientation, each node $\mathbf{x}_{GT, i}$ has its corresponding laser scan $\mathcal{P}_i$ and global descriptors $d_i$. In this research, we use only polar context descriptors.
• Figure 3: Loop closure detection between sessions. (a) the descriptors of each session data are compared against each other to find correspondences. (b) Some correctly and wrongly detected loop closures are shown in green and red, respectively. In blue, two encounters $c$ linked to the trajectories' respective anchors. The trajectory's offset concerning a common global frame is specified by the anchors $\Delta$.
• Figure 4: Pose graph optimization with multiple sessions. (a) In orange is the drifted map created by a SLAM system (exaggerated for illustrative purposes); in red is the respective trajectory; in green is the ground truth trajectory (b) Each pose graph optimization iteration tries to create a consistent global map placing the scans closer to the ground truth. (c) The final map is correctly aligned with the BIM model.
• Figure 5: Detected positive differences in the point cloud and the BIM model. (a) The original aligned corrected point cloud (b) Only voxelized clustered new objects in the scene. The ceiling was removed for better visualization.