Road Network Generation and Closed-Loop Optimization for Intelligent Transportation Systems

The generative design and optimization of road networks are crucial for intelligent transportation systems. However, existing automated methods that generate pixel-based road networks often result in suboptimal connectivity, leading to inefficient transportation. Furthermore, these methods seldom consider the variability in daily travel demand when optimizing road networks. To alleviate these problems, we introduce a new deep generative model and a topology-aware loss function to enhance road connectivity. Additionally, we propose an optimization method that adapts to daily travel demands. This method refines road networks through iterative feedback, resulting in more efficient and practical transportation solutions. Experimental results confirm the effectiveness of our approach in generating high-quality and traffic-efficient road networks.