Path Planning for Autonomous Robots Using a Large Language Model With Access to Local Building Models

In dynamically changing construction sites, the effective use of BIM and large language models (LLMs) is a key challenge. This study enables long-term reasoning by providing an LLM, which lacks persistent memory, with local memory. We propose an LLM-centered system integrating a database and BIM software, and demonstrate its capability to generate navigation paths for mobile agents within construction sites. Furthermore, motion simulations of the generated paths are conducted using ROS 2, confirming that the proposed system can serve as a novel approach for robot motion instruction in construction-site environments.