Maximum entropy inverse reinforcement learning for campus spatial optimization from Wi-Fi probe trajectories: a case study of Southeast University Wuxi Campus

Abstract With the outward expansion of university campuses toward suburban areas, the modern campus has evolved into an increasingly independent social space, providing an organizational setting in which users’ daily activities rely on diverse on-campus facilities. Accordingly, the rationality of spatial organization is a critical determinant of quality of life and a central focus of campus optimization research. However, existing methods for campus space optimization predominantly rely on the experience of planners and pedestrian flow simulation tools based on spatial topology, rule-based models, deep learning, or reinforcement learning, while often failing to adequately incorporate the influence of environmental factors on behavior. In reality, the behavior of campus users tends to be goal-oriented, and neglecting environmental factors may therefore lead to design biases. Inverse Reinforcement Learning (IRL) has the capacity to capture human interactions with various environmental factors in real-world settings, offering a novel approach to pedestrian trajectory simulation on campuses. Thus, the Southeast University Wuxi Campus serves as the research site, with pedestrian trajectory data and environmental feature data collected. A maximum entropy inverse reinforcement learning (MaxEnt IRL) method is employed to construct a pedestrian trajectory simulation model, which is subsequently validated and applied. Robustness and modeling rationality are further examined through environmental representation analysis and multi-resolution sensitivity experiments. Experimental results demonstrate that the proposed model effectively simulates pedestrian behavior under the influence of environmental factors and quantifies their relative attractiveness, thereby providing an effective tool for optimizing campus spatial organization.