When Curriculum Learning Meets Multi-Agent DRL in Connected Autonomous Vehicles

Efficient lane changing in Connected Autonomous Vehicles (CAVs) plays a pivotal role in congestion reduction by optimizing traffic flow, minimizing bottlenecks, and enhancing overall road capacity. This paper introduces an innovative automated lane-changing strategy based on Curriculum Learning, offering notable advantages in learning efficiency while maintaining stable performance. The trained agent acquires the capability to develop a safe and time-efficient driving policy for lane-change decisions. Numerical results provide a comprehensive perspective on the strengths of the proposed framework, opening avenues for future advancements in autonomous driving systems and intelligent transportation networks.