Multimodal ReAct Agent for Compliance Checking

This technical report presents a study on a multimodal ReAct agent designed for distance measurement in automated compliance checking (ACC) within building information modeling (BIM). The report outlines the challenges posed by ambiguous geometric measurements in regulatory documents and emphasizes the need for integrating graphical rules alongside textual regulations. The study introduces a prototype-level agent that utilizes six distance measurement types derived from accessibility regulations. It evaluates the agent's performance across different input conditions: graphical-only, textual-only, and combined inputs. The findings indicate that the use of combined inputs significantly enhances tool selection accuracy, with the Grok 4.1 model achieving the highest performance metrics. The report also discusses the methodology for implementing the agent, including dataset construction from regulatory frameworks such as the Americans with Disabilities Act (ADA) and the Act on the Promotion of Convenience for Persons with Disabilities in Korea. Overall, the study highlights the potential of multimodal approaches in improving compliance checking processes.