Resource Allocation for Cooperative Sensing in Fog Computing-Based Vehicular Networks

Cooperative sensing leverages the collaborative exchange of sensing and driving data among vehicles to broaden the sensing range and refine accuracy for autonomous vehicles. However, the substantial surge in shared data has engendered a significant demand for wireless spectrum in vehicular networks (VNETs), rendering the execution of driving safety tasks within tolerable latency more challenging under resource-constrained circumstances. In this study, a cooperative sensing scheme based on clusters in fog computing-based vehicular networks (VNETs) is proposed. Here, vehicles are grouped into collaborative clusters, selectively exchanging sensing messages to enable cooperative computing. To balance communication, sensing, and computing performance in cooperative sensing, a joint problem for vehicle clustering formulation, sensing message selection, radio, and computing resource allocation is established. This problem is subsequently decomposed into two sub-problems due to their non-convex nature. The vehicle clustering formulation alongside sensing message selection sub-problems are redefined into a multi-agent Markov game, addressed via an attention-enhanced multi-agent deep reinforcement learning algorithm. Additionally, the swap matching and interior-point methods are adapted for iterative optimization of sensing, communication, and computing performances. Comprehensive simulation results have validated the superiority of the proposed scheme over existing models across diverse performance metrics