Optimizing Urban Intersection Management in Mixed Traffic Using Deep Reinforcement Learning and Genetic Algorithms

This study aims to optimize lane configurations at urban intersections within mixed traffic environments, integrating both Connected Autonomous Vehicles (CAVs) and Human-driven Vehicles (HVs). By employing genetic algorithm and deep reinforcement learning (DRL), the research seeks to dynamically adjust lane configurations to improve intersection efficiency under varying traffic conditions and CAV penetration rates. This study utilizes a genetic algorithm to solve the dynamic lane configuration problem in a mixed traffic environment and concludes that higher traffic volumes require dedicated CAV lanes to significantly reduce vehicle delays. Additionally, the effectiveness of dynamic lane configuration is further validated through the DRL model, which shows significant improvements in average speed and waiting time as the CAV penetration rate increases. The study highlights the importance of adaptive strategies for managing complex urban traffic, providing valuable insights for future urban traffic management and planning. The integration of genetic algorithm with DRL underscores the potential for developing flexible and efficient solutions to optimize urban intersection management in mixed traffic environments. These findings suggest that adaptive lane configuration strategies can support the broader adoption of autonomous driving technologies and contribute to the development of smarter and more efficient urban transportation systems.