Co-movement aware trajectory generation via waypoint-guided generative adversarial networks

Abstract Synthetic trajectory generation is essential for addressing privacy concerns and data scarcity in mobility-related applications. Although existing solutions effectively capture general spatio-temporal features, they often overlook co-movement patterns among moving objects, which are crucial for applications such as traffic simulation, ride-sharing, and crowd modeling. Moreover, most approaches rely on road network representations, limiting generalization and failing to preserve fine-grained mobility trends. To tackle these challenges, we propose CA-Gen, a Co-movement Aware trajectory generation framework based on Generative Adversarial Networks (GANs). Instead of employing road vertex mapping, we introduce a hot grid-cell based trajectory representation to enhance robustness and generalization. To better simulate real-world co-movement patterns, we design a way-point guided search algorithm based on frequent subsequence mining. Extensive experiments on real-world datasets show that CA-Gen significantly outperforms existing SOTA methods, generating realistic trajectories that retain both individual mobility characteristics and co-movement trends, providing a privacy-preserving and high-fidelity solution for mobility analysis.