Building industrial metaverses requires cost-effective, labor-saving, and robust 3D reconstruction of site environments. We have proposed a method leveraging 360-degree cameras, a robust image matcher (GIM), and 3D Gaussian Splatting (3DGS). In this talk, we detail the processing workflow and demonstrate its robustness across diverse environments by comparing it with standard SfM/3DGS pipelines (e.g., nerfstudio). Furthermore, we quantitatively evaluate the rendering quality of the reconstructed models using metrics such as PSNR and SSIM to discuss the method's practicality for industrial applications.