Complexity Controllable Road Network Generation for Virtual Testing of Autonomous Driving

Complexity controllable road network generation is crucial for accelerating autonomous vehicle (AV) virtual simulation testing. This study proposes a generation method via optimized combination of realistic road elements: first, real-world urban road networks are decomposed into elements; high-collision-risk elements (e.g., T-junctions, merging/diverging zones) are abstracted into parameter-configurable graph models. These models are instantiated using real cartographic data, with complexity assessed by collision risk metrics and labeled via an evaluation function. A tunable optimization model selects diverse complexity elements to assemble non-intersecting, realistic, compact virtual road networks. Experimental validation with Xi’an cartographic data generated virtual road networks.