Learning Resilient Distributed Channel Access Policies in V2I Networks Under Intelligent Jamming

While the Internet of Vehicles (IoV) can revolutionize transportation systems through intelligent connectivity, a critical challenge in realizing this potential lies in ensuring efficient channel allocation in the IoV ecosystem, particularly considering dynamic channel conditions and adversarial jamming exacerbated by the emergence of Artificial Intelligence (AI)-based jamming. To address these challenges, in this study, we use Distributed Edge Intelligence (DEI) to propose a distributed channel access mechanism for the Vehicle-to-Infrastructure (V2I) mode of IoV networks. Specifically, using an actor-critic-based Multi-agent Reinforcement Learning (MARL) framework with a common critic, we model the distributed channel access problem in V2I communications under varying channel conditions and an intelligent jamming device (iJD) interference as a Decentralized Partially Observable Stochastic Game (Dec-POSG). Furthermore, by addressing challenges, such as partial observations, non-stationarity, and credit assignment, our proposed approach fosters collaboration among intelligent vehicles (iVs) without direct communication. In addition, our unique counterfactual reasoning-aided action evaluation mechanism and a novel utility function design enable the iVs to learn mixed collaborative-competitive channel access policies, thereby enhancing channel utilization, mitigating the impact of the, and improving the network's Sum Cross-Layer Achievable Rate (SCLAR).