From Automatic to Autonomous: a Large Language Model-driven Approach for Generic Building Compliance Checking

EasyChair Preprint 15364

Abstract

Despite extensive research and development in automating compliance checking (ACC) of building designs, a generic, scalable, and automated system remains elusive. Current ACC systems are limited to specific domains and rule types. The main challenges lie in the vast number of design requirements and the wide diversity of object concepts, attributes, constraints, and relations. This complexity demands an intelligent, autonomous system capable of independent planning, decision-making, and task execution to adapt to new situations, rather than traditional pre-programmed automation paradigms. Inspired by the agency capabilities of Large Language Models (LLMs) in solving complex tasks, this study explores their application in building compliance checking. The goal is to develop an LLM agent that can understand design requirements, plan the checking process, retrieve relevant BIM data, and execute checks autonomously. A proof-of-concept prototype was developed and an autonomous rule-checking experiment conducted with it has successfully demonstrated the potential of the LLM-driven approach.

Keyphrases: Artificial Intelligence Agent, Automatic Compliance Checking, Autonomous Compliance Checking, Building Information Modelling, Large Language Model

@booklet{EasyChair:15364,
  author    = {Huaquan Ying and Rafael Sacks},
  title     = {From Automatic to Autonomous: a Large Language Model-driven Approach for Generic Building Compliance Checking},
  howpublished = {EasyChair Preprint 15364},
  year      = {EasyChair, 2024}}