Integrating Preference-Aware Modeling of Human Spatial Behavior in Cyber-Physical-Human Systems

This study introduces a new approach for modeling preference-aware human spatial behavior using Graph Neural Networks (GNN) and Reinforcement Learning (RL). Current models often overlook the causality and impact of factors influencing preferences. Our approach utilizes GNN for its advanced handling of graph-structured spatial data, capturing physical, social, and environmental features and how these are perceived by humans. Integrated with RL, the model dynamically adapts to changes in the surrounding environment, improving adaptability and generalizability of simulations. As a proof of concept, we illustrate the approach in an educational conference room setting to compare student behavior simulation with and without preference inclusion. The results indicate that preference incorporation leads to significantly more realistic simulations, highlighting its potential to improve the design and control of cyber-physical-human systems.