AI-driven Computation Offloading in MEC-enabled Vehicular Networks: A Dual-Stage DRL Framework

As vehicular networks (VNs) increasingly rely on real-time applications such as autonomous driving and intelligent transportation systems (ITS), mobile edge computing (MEC) has become essential for reducing latency by offloading tasks to nearby servers. However, optimizing service delay in MEC-enabled networks is challenging due to the limited computational capacity of vehicles and a dynamic environment. This paper extends a dual-stage deep reinforcement learning (DRL) approach by integrating advanced DRL algorithms—prioritized experience replay DDPG (PER-DDPG), combined experience replay DDPG (CER-DDPG), twin delayed DDPG (TD3), and proximal policy optimization (PPO)—for power optimization and reduced propagation delays. Additionally, we optimize the core allocation, where we implement Dueling Double DQN (DDDQN) and Double DQN (DDQN). Simulation results show that DDDQN achieves the most efficient core allocation, leading to lower computational delays, highlighting the potential of advanced DRL techniques to enhance resource-constrained VNs and improve MEC systems for scalable ITS solutions.