Machine learning in pedestrian and evacuation dynamics for the built environment: A systematic literature review

The study of pedestrian and crowd movement has produced a plethora of publications over the past decades. Numerous knowledge-based models have been developed to describe, analyze and predict human motion behavior, particularly with respect to evacuation analysis to ensure public safety. In recent years, Machine Learning (ML) models have become widely successful across many disciplines, including applications for human behavior in the built environment, city planning, robotics and autonomous driving. In this review article, based on a systematic search of the Scopus database (2022–2024), we present a comprehensive overview of ML-based pedestrian and crowd models, highlighting the most popular approaches, as well as modern data collection methods that have led to public benchmark datasets and increasingly standardized model validation techniques. We analyze ML models that provide insights into crowd movement and evacuation performance, potentially supporting building design and safety assessment in the built environment, while outlining similarities and differences between these models with regards to behavioral traits such as goal-driven behavior and collision avoidance. Moreover, we review the involved learning paradigms, including supervised and reinforcement learning, and the associated quantities of interest that can be predicted, such as velocity, density, flow, and evacuation time.