Perception-Responsive Adaptive Virtual Reality for Construction Hazard Recognition: A Feasibility Study

Abstract

Hazard recognition deficiencies remain a primary factor contributing to accidents in the construction industry, where workers often fail to visually detect critical safety cues due to perceptual and attentional limitations. Although virtual reality (VR) has emerged as an effective tool for immersive safety training, most current systems do not account for individual differences in visual attention or perceptual capability. This study investigates the feasibility of a gaze-driven adaptive VR training framework that dynamically modifies instructional support and scenario complexity in response to trainees’ real-time perceptual performance. The system continuously captures eye-tracking metrics across predefined hazard Areas of Interest, including Time-to-First Fixation, fixation count, normalized dwell time, and scanpath dispersion. These metrics are synthesised into discrete perceptual performance states using a Support Vector Machine (SVM) classifier, which subsequently guides adaptive adjustments to hazard salience, visual guidance, and task difficulty within the VR environment. A pilot study involving 17 participants was conducted across baseline (n=7) and adaptive VR (n=10) conditions to evaluate system feasibility, real-time operation, and user experience. Results indicate that the framework effectively distinguished variability in participants’ perceptual patterns and successfully triggered adaptive modifications during training. The classifier achieved an overall accuracy of 88.3% with a ROC–AUC of 0.92, while usability evaluation using the System Usability Scale (SUS) produced a mean score of 84.25 for the adaptive condition, indicating strong perceived usability at the prototype stage. Overall, the findings demonstrate the feasibility of integrating gaze-based perceptual monitoring with adaptive VR safety training and highlight the potential of perception-responsive training systems for supporting construction hazard recognition.