Rapid scan-to-building information modeling using robotics and artificial intelligence for construction applications

摘要

The current industry practice for Building Information Modeling (BIM) relies on static laser scanning and manual annotation of data, which is labor-heavy and time-consuming. This research introduces robotics and artificial intelligence techniques to automate and expedite the scan-to-BIM process. A ground robot is used to collect laser scans from the construction site and create 3D point clouds that are automatically registered. A learning approach is used to train a navigation algorithm such that the robot is able to optimize scan efficiency without having a prior map. An embedding-based segmentation algorithm is then applied to parse and organize the raw data into semantically coherent entities. Finally, nearest-neighbor matching of solid 3D BIM objects is performed using a pre-built database of building elements. The proposed navigation algorithm led to a 17% reduction in scanning distance whereas the proposed segmentation algorithm led to a 9% improvement in the AMI score.