3D Model Generation from Open Data

Abstract

In recent years, large language models (LLMs) have advanced remarkably, and agentic AI systems utilizing the Model Context Protocol (MCP) to integrate diverse data sources and tools have become increasingly available. In order to generate three-dimensional models from incomplete data, it is necessary to construct a spatial representation framework that incorporates domain-specific knowledge from civil engineering. In parallel, it is also essential to systematize the data required to reconstruct the three-dimensional geometry of target structures and to develop methods for embedding this spatial representation framework into LLMs, enabling the generation of three-dimensional shape data. In this paper, we organize approaches for data integration and knowledge utilization based on MCP, propose a framework for constructing 3D models from incomplete data, and demonstrate an automated method for reconstructing three-dimensional shapes even in cases where data are insufficient or partially missing.