BIM Informed Visual SLAM for Construction Environments

Asier Bikandi-Noya; Miguel Fernandez-Cortizas; Muhammad Shaheer; Ali Tourani; Holger Voos; Jose Luis Sanchez-Lopez

cs.ROSep 17, 2025

BIM Informed Visual SLAM for Construction Environments

Authors: Asier Bikandi-Noya, Miguel Fernandez-Cortizas, Muhammad Shaheer, Ali Tourani, Holger Voos, Jose Luis Sanchez-Lopez

Organizations: Automation and Robotics Research Group, Interdisciplinary Centre for Security, Reliability, and Trust (SnT), University of Luxembourg, Luxembourg

Abstract

Monitoring building construction sites requires comparing the as-planned design with the as-built state, which can be estimated in real time using Simultaneous Localization and Mapping (SLAM) techniques. However, visual SLAM is prone to trajectory drift in construction environments, producing maps that are geometrically inaccurate with the actual environment. To address this limitation, we augment an existing RGB-D SLAM system with structural priors derived from the Building Information Model (BIM). The system associates detected walls with their BIM counterparts and includes these correspondences as geometric constraints in the back-end optimization, reducing drift and enhancing global consistency. The proposed method operates in real time and is validated on multiple real construction sites, achieving an average trajectory error reduction of 25.23% and a 7.14% improvement in map accuracy over state-of-the-art baselines. Robustness analyses further demonstrate resilience to incomplete BIM data and geometric discrepancies between as-planned models and the as-built environment.

Figures & tables

Fig. 1 : Proposed BIM-informed RGB-D SLAM system. Detected as-built walls are matched to their as-planned BIM counterparts, with a 2D projection of the alignment.

Fig. 2 : System architecture of ivS-Graphs. The pipeline takes BIM and RGB-D camera data as inputs and BIM walls are extracted offline from the BIM model. The SLAM backbone front-end processes visual data into keyframes, map points, and wall segments. Our contributions are highlighted in green: ( 1 ) initial alignment followed by a ( 2 ) continuous wall association and ( 3 ) back-end graph optimization incorporating wall-to-wall constraints ( $\mathbf{W_{\text{BIM}}}\leftrightarrow\mathbf{W_{S}}$ ). This constrains camera drift to produce an optimized as-built map (right) aligned with the design.

Env.	Sequence	#W	#S	Area ( $\mathbf{m}^{2}$ )	L (m)	D(s)
Offices	office1-1	10	4	100	90	670
	office1-2	12	3	109	48	265
	office1-3	12	4	113	99	524
	office1-4	6	2	54	37	169
	office2-1	10	3	168	68	268
	office2-2	10	3	168	73	370

TABLE I : Summary of evaluation sequences. #W and #S denote the number of visible walls and distinct navigable spaces (rooms and corridors), respectively. Area, Length (

L

), and Duration (

D

) are in

m^{2}

m

, and

s

. CS1–CS3-Furn are distinct construction sites at different completion stages; office1 and office2 are separate buildings with multiple sequences each.

Param.	Value	Description	Section
$\tau_{\perp}$	0.17 rad	Perpendicularity threshold for initialization	III-C
$\tau_{init}$	0.30 m	Max. residual for initial alignment	III-C
$\tau_{p}$	0.77 m	Plane-parameter distance threshold	III-C
$\tau_{c}$	6.00 m	Lateral centroid distance threshold	III-C
$\alpha$	0.70	Weighting for plane vs. spatial proximity	III-C
$\beta$	$10^{-2}$	Base weight for wall-to-wall factors	III-D

TABLE II : Key Parameters and Thresholds for ivS-Graphs

Sequence	ATE ( $\mathrm{m}$ )						Prec. (%) (ours)
Sequence	BAD SLAM [ 8 ]	ORB- SLAM3 [ 7 ]	DROID- SLAM [ 9 ]	vS-Graphs [ 6 ]	ivS-Graphs (ours)	Diff. (%)	Prec. (%) (ours)
office1-1	-	0.462	0.541	0.343	0.226	+33.99%	90
office1-2	1.653	0.365	0.533	0.353	0.183	+48.15%	100
office1-3	-	0.272	0.360	0.253	0.175	+30.83%	91
office1-4	-	0.101	0.109	0.090	0.120	-33.33%	100
office2-1	1.879	0.160	0.395	0.223	0.214	+4.04%	100

TABLE III : ATE ( ATE ) in meters (

\mathrm{m}

) for the evaluated VSLAM algorithms. The best and second-best results are boldfaced and underlined , respectively. Values are averaged over five runs. The minimum completion rate of each sequence is set to 50%, and dashes indicate unavailable data due to tracking failure. The Diff. column reports the improvement of ivS-Graphs over vS-Graphs baseline , and final column Prec. (%) reports the wall-matching precision .

Fig. 3 : Accumulated trajectory drift error over time in office1-2 sequence, showing the ATE for baselines and ivS-Graphs .

Fig. 4 : Top-down overlay trajectory overlay on office1-1 . Left: ivS-Graphs (with BIM). Right: vS-Graphs (init. align. only)

Sequence	Point Cloud RMSE ( $\mathrm{m}$ )
Sequence	BAD SLAM [ 8 ]	ORB- SLAM 3 [ 7 ]	DROID- SLAM [ 9 ]	vS-Graphs [ 6 ]	ivS-Graphs (ours)	Diff. (%)
office1-1	-	0.416	0.430	0.403	0.331	+17.88%
office1-2	0.442	0.537	0.435	0.444	0.396	+10.72%
office1-3	-	0.360	0.395	0.340	0.354	-4.15%
office1-4	-	0.349	0.323	0.327	0.306	+6.35%
office2-1	0.394	0.407	0.371	0.432	0.382	+11.55%

TABLE IV : Root Mean Square Error (RMSE) in meters of reconstructed point clouds for the evaluated VSLAM algorithms. The best and second-best results are boldfaced and underlined , respectively. Values are averaged over 5 runs. The minimum completion rate of each sequence is set to 50%, and dashes indicate missing data due to tracking failure. The final column reports the improvement of ivS-Graphs over vS-Graphs .

Sequence	Nominal	20% Miss.	30% Miss.	Rel. Err. (%)
office1-1	0.226	0.244	0.253	+11.9%
CS1	0.100	0.104	0.094	-6.0%
CS2-Mid	0.156	0.188	0.166	+6.4%
CS3-Furn	0.149	0.157	0.162	+8.7%
Mean	0.158	0.173	0.169	+5.3%

TABLE V : ATE [m] under varying BIM completeness. Nominal uses the full model; 20% and 30% Miss. omit walls. Rel. Err. shows ATE increase from Nominal to 30% Missing case.

Seq.	Nom.	Moderate	Severe	Rel. Err. (%)
Seq.	ATE	ATE / MR	ATE / MR	Mod.	Sev.
off1-1	0.23	0.29 $\pm$ 0.05 / 96.0 $\pm$ 5.5	0.30 $\pm$ 0.16 / 78.0 $\pm$ 8.4	+28.3	+33.4
CS1	0.10	0.16 $\pm$ 0.06 / 91.3 $\pm$ 9.5	0.16 $\pm$ 0.04 / 85.0 $\pm$ 12.2	+63.4	+64.1
CS2-Mid	0.16	0.18 $\pm$ 0.04 / 80.0 $\pm$ 13.9	0.22 $\pm$ 0.09 / 73.3 $\pm$ 22.4	+12.5	+38.0
CS3-Furn	0.15	0.18 $\pm$ 0.08 / 94.5 $\pm$ 5.0	0.40 $\pm$ 0.26 / 81.8 $\pm$ 6.4	+20.2	+167.7
Mean	0.16	0.20 / 90.4	0.27 / 79.5	+28.0	+71.1

TABLE VI : ATE [m] and Match Rate (MR %) under geometric deviations, reported as ATE / MR. Moderate (

\pm 0.2\,\text{m}

\pm 5^{\circ}

) and Severe (

\pm 0.8\,\text{m}

\pm 15^{\circ}

) perturbations, averaged over 5 randomized runs (

\pm

standard deviation), are applied to the BIM walls. Rel. Err. indicates ATE % increase over Nominal, whose MR is 100% for all sequences.

Sequence	vS-Graphs	ivS-Graphs (Ours)
Sequence	FPS [6]	FPS	Align. (ms)	Match (ms)	Opt. (ms)	Opt. calls
office1-1	23.8	21.8	0.24	0.06	843	67
CS1	23.3	22.5	0.23	0.07	542	38
CS2-Mid	24.2	23.9	0.34	0.06	321	22
CS3-Furn	25.5	24.9	0.15	0.05	442	18
Mean	24.2	23.3	0.24	0.06	542	36

TABLE VII : Runtime performance. FPS is the average processing rate over the trajectory. Align. is the one-time initial registration cost. Match and Opt. (ms) are the mean per-call costs of continuous wall matching (per detected wall) and BIM back-end optimization.

Explore similar work

MVP-SLAM: Multi-Camera Visual-Inertial Floorplan-Prior SLAM
Sep 30, 2026Asier Bikandi-Noya, Miguel Fernandez-Cortizas, Muhammad Shaheer +2Orb-Slam2 Simultaneous Localization And Mapping
Authors are with the Automation and Robotics Research Group, Interdisciplinary Centre for Security, Reliability, and Trust (SnT), University of Luxembourg, Luxembourg. · Holger Voos is also associated with the Faculty of Science, Technology, and Medicine, University of Luxembourg, Luxembourg.
Hilti-Trimble-Oxford Dataset: 360 Visual-Inertial Benchmark with Floor Plan Priors for SLAM and Localization
Jul 7, 2026Samuele Centanni, Yuhao Zhang, Yifu Tao +8Simultaneous Localization And Mapping Visual-Inertial Systems
Hilti AG, Corporate Research & Technology, Schaan, Liechtenstein · University of Oxford, Dynamic Robot Systems Group, Oxford, UK · ETH Zürich, Robotics Systems Lab, Zürich, Switzerland +2
BIM-Loc: BIM-Integrated Discrepancy-Aware LiDAR-based Indoor Localization
Jun 12, 2026Yinqiang Zhang, Liang Lu, Yipeng Pan +5Indoor Localization Discrepancy
Department of AI & Data Science, University of Hong Kong (HKU), Hong Kong, China. · Humanoids and Human Centered Mechatronics Research Line, Italian Institute of Technology (IIT) Genoa, Italy. · Department of Architecture and Civil Engineering, City University of Hong Kong (CityU), Hong Kong, China.

Sep 30, 2026, cs.ROJ/K move · Enter open · S save

MVP-SLAM: Multi-Camera Visual-Inertial Floorplan-Prior SLAM

Asier Bikandi-Noya, Miguel Fernandez-Cortizas, Muhammad Shaheer, Holger Voos+1

Authors are with the Automation and Robotics Research Group, Interdisciplinary Centre for Security, Reliability, and Trust (SnT), University of Luxembourg, Luxembourg. · Holger Voos is also associated with the Faculty of Science, Technology, and Medicine, University of Luxembourg, Luxembourg.

Orb-Slam2 Simultaneous Localization And Mapping Floor Plan Generation

Open paper AskPDF arXiv

Indoor building construction sites are demanding environments for visual SLAM, where variable lighting and repetitive, low-textured structures make the system drift over long trajectories, though structural elements such as walls remain distinguishable despite these conditions. These buildings are constructed according to their as-planned floor plans, available from the design phase, and although the actual as-built site can differ from this design, floor plans still provide a metric reference, both to localize the system in the building and to correct drift. Existing methods often use the floor plan to correct an already-built trajectory offline, and those that instead correct it online typically rely on depth sensors. We instead present MVP-SLAM, an online visual-inertial SLAM on two opposite-facing fisheye cameras that corrects drift from cameras alone by matching walls detected in its map to the floor plan, through a drift-aware policy. A multi-stage integration then turns each matched pair incrementally into a persistent correction, so the trajectory stays corrected and localized within the floor plan as it is built. MVP-SLAM was validated on the multi-floor construction sites of the Hilti-Trimble SLAM Challenge 2026, ranking 2nd of 22 teams in the Localization task (0.29 m mean RMSE) and 5th of 62 teams in the SLAM task (0.24 m), the top-ranked one in both tasks among those that operate online, integrate the floor plan, and localize within it.

Abstract

Explore similar work

MVP-SLAM: Multi-Camera Visual-Inertial Floorplan-Prior SLAM

Hilti-Trimble-Oxford Dataset: 360 Visual-Inertial Benchmark with Floor Plan Priors for SLAM and Localization

BIM-Loc: BIM-Integrated Discrepancy-Aware LiDAR-based Indoor Localization