Based on Hierarchical Reinforcement Learning for Large-scale Pedestrian Trajectory Prediction

Abstract

This paper describes the construction of an airport terminal simulation model using AnyLogic simulation software. It considers the advantages of hierarchical reinforcement learning and divides the complete process of pedestrian trajectories at the airport into layers. Pedestrians are treated as intelligent agents for hierarchical reinforcement learning. A large-scale pedestrian trajectory planning algorithm based on hierarchical reinforcement learning is designed to match the hotspots in the airport region simulated by pedestrian trajectories with congested areas in the terminal scene. A comparison is made with traditional multi-agent Q-learning algorithms and single-table hierarchical reinforcement learning. The results show that our algorithm can accurately identify the pedestrian flow hotspots in the actual terminal, with improved matching accuracy compared to traditional multi-agent Q-learning algorithms and single-table hierarchical reinforcement learning. The algorithm also exhibits faster convergence speed.