Analysis and Modeling of Worker Trust in Automated Guided Vehicles for Manufacturing Workplace

Abstract

Despite the prevalence of leveraging Automated Guided Vehicles (AGVs) to streamline transportation on modern manufacturing floors, they typically operate separately from human workers, partially due to people’s lack of trust. To reach their full potential, AGVs must be integrated into manufacturing environments alongside humans, where they can sense and respond to workers’ trust levels. Research has shown that performance-based factors, such as autonomous agents’ approaching direction and speed, significantly contribute to developing human trust in such systems. However, the results of these studies cannot be easily generalized to the human-AGV interaction due to the context-dependent nature of human trust. Thus, a 2 (AGV Deceleration Rate) × 8 (AGV Approaching Direction) × 2 (User’s Expected Crossing Path) simulation study was conducted to examine the trust dynamics of workers when interacting with AGVs. Analysis of Variance revealed that the AGV’s approaching direction, the user’s expected crossing path, and their interaction significantly impacted the user’s trust. Machine learning models were trained on behavioral and physiological signals to predict trust, including linear regression, bagging, and eXtreme Gradient Boosting (XGBoost) regressor. Among the models tested, XGBoost regressor achieved the best performance, with an R² of .65 and an RMSE of 1.03. These findings highlight key design considerations for developing trust-aware AGVs as essential steps toward human-centered industrial settings.