Reinforcement Learning for Crowd Control Optimization with An A* Based Social Force Interaction Model

During the process of modern urbanization, the number of large public places with a high density of crowd is also increasing. To cope with the high level of congestion and security threats in crowded scenarios such as subway stations during rush hours, firstly, this paper introduce a social-force interaction model based on A* algorithm (ASIM), which can simulate the irrational behavior of pedestrians in crowd scenarios and evaluate the effectiveness of the crowd control strategies. Secondly, based on ASIM, we present a reinforcement learning algorithm based on the proximal policy optimization (PPO) method, which can optimize the crowd control policies in various scenarios. Third, we integrate these two models into a unified framework called Crowd-Flow (CFF). The workflow based on this framework can easily simulate scenarios and give real-time crowd control strategies for different optimization scenarios and goals. Finally, we demonstrate the performance of CFF on a real-life subway station crowd control scenario. Compared with traditional strategies, CFF shows promising results in reducing crowd congestion.