Prof. Jürgen Melzner presented the research contribution entitled:

“Bridging the Safety Gap: Feasibility of Using Large Language Models for Hazard Detection in Robotic-Aided Construction Site Monitoring”

Authors: Ronny Helbing, Nasim Babazadeh, Prof. Dr. Jürgen Melzner (Bauhaus-Universität Weimar) & Dr. Chris Jurgens (Stellenbosch University)

The study focuses on the potential of multimodal Large Language Models (LLMs) to support the automated detection of safety risks on construction sites. The motivation stems from the persistently high number of fall-related accidents and incidents involving falling objects, which, according to current statistics, remain among the most frequent and severe causes of harm in the construction industry.

Using real construction site imagery, five state-of-the-art models were evaluated with regard to their ability to identify and assess hazards in edge protection systems. The results highlight the clear potential of AI-based image analysis for robotic safety inspections, while also revealing current limitations in terms of reliability and output consistency. These findings provide valuable impulses for future research aimed at advancing AI-supported safety monitoring systems on construction sites.

The Chair would like to thank the organizers of ACCCBE and CONVR25 for an inspiring conference and the valuable international exchange at the intersection of digitalization, robotics, and construction site safety.