Decentralized Multi-layer Vehicular Edge Computing Framework for Time-Efficient Task Coordination

Vehicle edge computing (VEC) offers substantial potential to enhance real-time applications in the connected autonomous vehicle (CAV) sector by positioning edge servers near CAVs. The need to balance real-time application demands and the diversity of tasks underscores the criticality of carefully deciding whether to process these tasks locally or to offload them for remote processing. This decision is pivotal in achieving efficient resource management, particularly in dynamic vehicular environments, emphasizing the requirement of exquisite policies to optimize VEC systems’ performance within the CAV ecosystem. While centralized policies face limitations in managing largescale deployments, decentralized approaches often fail to foster effective collaboration and coordination among CAVs. This paper proposes a novel multi-layered, decentralized framework designed for time-efficient task management and optimized network resource utilization in dynamic vehicular scenarios to address these limitations. This framework incorporates a novel multiagent deep reinforcement learning (MADRL) algorithm to enable intelligent decision-making for efficient task execution. The proposed algorithm considers the diverse computing capabilities of network entities, aiming to reduce latency without compromising task completion rates. Simulation results demonstrate that our proposed algorithm has superior performance in latency reduction while improving task completion rates compared to existing algorithms.