M-MADDPG: Research on Cooperative Optimization of Task Offloading and Network Resources in Vehicular Networks

Efficient resource allocation and task offloading remain critical challenges in intelligent transportation systems. This letter introduces M-MADDPG, a multi-agent deep reinforcement learning algorithm enhanced with multi-head attention, to jointly optimize task offloading, transmission power, and computing frequency in dynamic vehicular environments. The proposed attention mechanism significantly improves feature extraction and agent cooperation. Extensive simulations demonstrate that M-MADDPG consistently outperforms conventional methods in throughput and adaptability, providing a robust and scalable solution for future vehicular networks.