VESTA-MADDRL: Joint Task Offloading and Service Placement in MEC-Vehicle Networks Based on Improved MADDPG

Mobile Edge Computing (MEC) addresses latency issues by deploying resources at the network edge. However, integrating MEC with mobile vehicles introduces challenges including mixed discrete-continuous resource allocation, dynamic offloading conflicts, and limited vehicle storage constraints. This paper proposes VESTA-MADDRL, a novel framework for joint task offloading and service placement in MEC-empowered mobile-vehicle integrated networks. VESTA-MADDRL employs three core technologies to solve these challenges. First, it adopts an enhanced multi-agent deep deterministic policy gradient algorithm to handle mixed discrete-continuous action spaces, achieving unified optimization of binary offloading decisions and continuous resource allocation. Second, it utilizes constraint-aware action mapping with combinatorial selection mechanisms to resolve resource conflicts and ensure feasibility through Q-value-based knapsack optimization. Third, it implements hierarchical action decomposition with probabilistic rounding and greedy repair mechanisms to satisfy vehicle storage constraints while maintaining end-to-end differentiability. Experimental results demonstrate that VESTA-MADDRL achieves superior convergence and significantly outperforms baseline approaches in energy efficiency and latency reduction.