Hogan, RicardoRicardoHoganHsu, Wei-YiWei-YiHsuJACOB JE-CHIAN LINLiang, Ci-JyunCi-JyunLiang2025-07-312025-07-312025-0908873801https://www.scopus.com/record/display.uri?eid=2-s2.0-105007927817&origin=resultslisthttps://scholars.lib.ntu.edu.tw/handle/123456789/730854Unmanned ground vehicles (UGVs) have emerged for construction site geometric and visual data collection, which has traditionally been resource-intensive and hazardous. The use of multiple small UGVs can overcome single large UGV challenges such as coverage or collision hazards. The swarm simultaneous localization and mapping (SLAM) algorithm enables multiple UGVs to share data simultaneously, but it has several limitations, especially at the beginning when loop closure is not detected. This research introduces a model-driven UGV swarm system to address loop closure limitations for automated visual data collection. The system allows UGV swarms to autonomously navigate and gather visual and motion data, which are then refined by the swarm SLAM front end and sent to a base station. At the base station, the Sswarm SLAM back end computes loop closures between the UGVs while also detecting additional loop constraints with the building information modeling (BIM) approach to generate accurate point cloud data. The experimental results demonstrated that the proposed framework in the "two robots with BIM"case improved the point cloud accuracy by 15% and reduced the capture time by 35% compared to the "single robot without BIM"case. These findings highlight that the integration of the BIM model and swarm SLAM technology can enhance construction site data collection processes.false[SDGs]SDG9[SDGs]SDG17journal article10.1061/JCCEE5.CPENG-66192-s2.0-105007927817