Towards agent : LLM-based framework for robotic construction by leveraging IFC

Abstract

This work explores the integration of Large Language Model (LLM)-powered agents into robotic construction workflows. An LLM-based agent framework by leveraging standardized IFC format is proposed to guide robotic tasks, such as bricklaying and 3D concrete printing. Through a combination of tailored tools, prompt templates, and a CustomMemory module, the framework enables efficient extraction of building information and execution of robotic tasks. Finally, experimental validation in a simulated environment demonstrates the framework's potential to streamline robotic construction workflows from textual or voice commands to physical construction activities, while reducing redundant operations and enhancing reliability through thoughtful design.