Enhancing Construction Site Digital Twins with Language Embedded Radiance Fields

This study presents an innovative approach enhancing construction site digital twins by integrating neural radiance fields (NeRF) with language embedded radiance fields (LERF). This combination enables high-fidelity 3D reconstruction with natural language interaction capabilities. NeRF captures detailed spatial data from standard photographs, while LERF incorporates semantic understanding through natural language processing. Our implementation at Purdue University’s S-BRITE Center demonstrates the feasibility of accurately identifying construction materials, structural elements, and environmental conditions across diverse lighting scenarios. While current computational requirements and technical complexity present adoption challenges, this research establishes proof of concept that such systems can be developed and optimized. The results confirm significant future opportunities for site documentation, monitoring, and decision-making processes as computational resources advance. This research establishes the transformative capability of NeRF-LERF integration to revolutionize construction management practices.