VoDE-Crowd: Differential Evolution Optimized Crowd Simulation Using Voronoi Diagrams

Multi-layer crowd simulation frameworks often produce trajectory distortions and polygonal paths near obstacles, as the interaction between the global cognitive (navigation) and local physical layers deviates from natural pedestrian motion. To address this issue, this paper proposes VoDE-Crowd, a crowd simulation method that embeds an attraction-based Voronoi boundary field into the physical layer and employs a differential evolution algorithm to optimize the structure of the navigation field in a data-driven manner. The proposed Voronoi field guides agents along potential energy valleys defined by cell boundaries, while differential evolution directly searches for Voronoi generator points that improve the similarity between simulated and real pedestrian trajectories. Unlike existing methods that tune parameters within a fixed structure, our approach focuses on structural optimization of the navigation field itself, thereby guiding the simulation to produce smoother and more realistic trajectories. Experimental validation in a real-world airport scenario demonstrates that VoDE-Crowd outperforms conventional multi-layer simulation models in trajectory similarity and motion smoothness, confirming its effectiveness for data-driven crowd simulation in large-scale environments.