A Hierarchical Road Network Topology Generation Method Based on Graph CVAE and Graph Transformer

Road network generation plays a crucial role in urban planning and design. Compared to traditional procedural modeling, road network generation techniques based on generative artificial intelligence significantly reduce the amount of manual operations and cumbersome parameter settings. However, this approach often neglects the consideration of road hierarchy when generating road networks. Properly classifying the generated road networks can not only more accurately reflect the actual road layout but also provide essential data support for autonomous driving simulation tests. In light of this, this paper focuses on employing graph-based generative techniques to construct hierarchical road network topologies. We innovatively propose a novel generative model that combines Graph Conditional Variational Autoencoders (GraphCVAE) with Transformers, capable of creating hierarchical road network topologies that meet specific conditions. The effectiveness of this model is validated through a newly created dataset encompassing road networks from 71 Chinese cities, ensuring broad applicability of the generated topologies.