Scalable Hybrid-Quantum MARL for Low-Latency V2V Resource Allocation in Dense Urban Scenarios

Efficient resource allocation is critical for low-latency vehicle-to-vehicle (V2V) communications in dense urban environments. However, conventional multi-agent deep reinforcement learning (MARL) methods suffer from scalability issues due to large state-action spaces and heavy computational complexity. To address this, we propose a hybrid-quantum MARL (HQMARL) framework, where hybrid quantum neural networks (HQNN) replace parts of deep neural networks with parameterized quantum circuits. HQMARL preserves decentralized decision-making while reducing the number of trainable parameters and accelerating convergence. Simulation results show that HQMARL achieves over a 93% reduction in parameters (approx. 12.4k vs. 180k) and noticeably faster convergence speed (reaching stability in roughly half the iterations) compared to classical MARL, while maintaining high throughput and latency satisfaction in large-scale V2V scenarios.