Generative Transformer Model Synthesizes Human Mobility Patterns

A new generative model for synthesizing human mobility patterns has been proposed in a paper on arXiv (2405.17468). The model, a Transformer-based architecture, addresses limitations in existing deep learning and activity-based approaches by incorporating socio-demographic and household attributes to generate daily activity chains and complete trajectories. It is trained on open-source household travel survey data and fine-tuned with local data, demonstrating transferability to California, Washington, and Mexico. The research aims to improve transportation, urban planning, and public health applications by providing a cost-effective and adaptable alternative to traditional models.