Proactive Self-Aware Multi-Agent Reinforcement Learning for Resource Management in NR-V2X Platooning

This paper investigates the problem of resource management in NR-V2X (New Radio Vehicle-to-Everything) platooning, addressing the critical trade-off between Age of Information (AoI), Cooperative Awareness Messages (CAMs) demand and finite on-board energy reserve that traditional methods ignore. We propose a novel, Self-sufficient, Energy-Aware, Multi-agent Deep Deterministic Policy Gradient (SEAM-DDPG). By allowing platoons to enter low-power sleep mode strategically to conserve energy for future transmissions. To enhance the robustness of agents under the Centralized Training, Decentralized Execution (CTDE) paradigm, we transform agents from being reactive to proactive by incorporating a one-step forward model and imagination-augmented policy. Our simulation results demonstrate that SEAM-DDPG significantly prolongs the operational life of platoons, while maintaining communication performance. Specifically, SEAM-DDPG (Proactive) achieves a remarkable 94.59% improvement in average energy reserve and a 42.83% increase in operational duration compared to our baselines. This achieves significant 568.39% increase in final energy reserves over a simulation. It also reduces average power consumption by 30.76%.