Quantum multi-agent deep reinforcement learning for energy-efficient vehicular networks

In this paper, we address the complex mixed-integer nonlinear programming problem associated with channel assignment and joint power-energy allocation in urban platoon-based cellular-vehicle-to-everything (C-V2X) networks. In this context, the potential advantages of integrating quantum neural networks (QNNs) with classical multi-agent deep reinforcement learning (MADRL) approaches are investigated. Specifically, we combine a variational quantum circuit (VQC) with traditional neural networks to develop a hybrid quantum-classical neural network for the MADRL training process. Our goal is to employ this hybrid quantum-classical approach to simultaneously minimise the average age of information (AoI) which quantifies the freshness of information exchange between vehicle platoons and the roadside unit (RSU), maximise the cooperative awareness message (CAM) exchange probability among vehicles within the same platoon, and foster sustainable, green communication strategies through efficient management for both power and energy. We introduce the innovative decomposed multi-agent deep deterministic policy gradient (DE-MADDPG) algorithm, which is integrated with the twin delayed deep deterministic policy gradient (TD3) technique and advanced quantum computing technologies, resulting in our proposed hybrid quantum-classical decomposed multi-agent TD3 (DE-MATD3) algorithm. Compared with classical approaches, our numerical results reveal that the proposed algorithm achieves exceptional energy efficiency performance, while maintaining the algorithm convergence rate and AoI levels.