Mechanism Design-Based Multi-Agent Reinforcement Learning for Communication Resource Allocation in Cooperative Adaptive Cruise Control

Efficient resource allocation in Cooperative Adaptive Cruise Control (CACC) communication is crucial for maintaining reliable and effective data transmission between connected vehicles. However, existing decentralized optimization methods often lead to inefficient resource utilization, as agents prioritize individual returns over global system welfare. This paper introduces a novel decentralized training framework that leverages counterfactual information to evaluate the potential impact of each agent’s actions on the global system objective. By modifying reward signals based on this evaluation, our approach steers individual strategies toward cooperative behaviors, enhancing overall system performance. Simulation results demonstrate that our mechanism design-based MARL framework outperforms conventional decentralized methods, ensuring both effective resource allocation and reliable string stability in complex CACC scenarios.