Modelling emergent pedestrian evacuation behaviours from intelligent, game-playing agents

Pedestrian evacuation modelling plays a crucial role in emergency planning and management, yet existing models often inadequately reproduce the complexity of human evacuation behaviours, particularly the dynamic interactions and anticipatory decision-making among individuals during evacuations. Traditional pedestrian evacuation models typically rely on simply or predetermined behaviours. They generally underestimate people’s intelligence in real-world emergencies, failing to realistically replicate emergent patters of individual evacuation behaviours, especially in highly constrained and densely populated spaces. This limitation hinders models’ simulation accuracy for evacuation dynamics and also undermines their reliability for evacuation planning and crowd management.
Here, an innovative agent-based model incorporating Bayesian Nash Equilibrium (BNE) has been introduced to address this gap. The approach integrates game-theoretic decision-making process within individual agents, enabling them to dynamically predict crowd congestions and avoid potential dangers. This thesis comprises three peer-reviewed studies that evaluate the performance of the BNE-informed model against naïve behavioural models (e.g. Shortest Route (SR) and Random Follow (RF) models) across varied evacuation contexts, from simplistic tunnel environment to the real-world layout of Grand Central Terminal (NYC). Results indicate that intelligent, game-playing agents are capable to evacuate with less exit time and higher personal comfort by intelligently navigating away from congested areas and avoiding potential threats. It was also found that the proposed models can reproduce emergent patterns of individual evacuation behaviours, such as herding, self-organized queuing, and outsmarting behaviours especially evident in the scenarios with high density and extremely limited spatial alternatives.
These findings conclusively demonstrate that incorporating Bayesian game theory into pedestrian evacuation modelling can significantly enhance their simulation accuracy. This BNE-informed model not only fills a crucial gap in evacuation modelling literature but also provides a robust framework for evacuation planning and management. The broader significance of this research lies in its potential to improve real-world evacuation strategies and enhance the management of large-scale emergency scenarios.