Enhancing task offloading in vehicular networks: A multi-agent cloud-edge-device framework

Abstract

In vehicular networks, the increasing demand for computational resources often exceeds the capabilities of in-vehicle devices. To address these challenges, we propose a cloud-edge-device collaborative framework integrated with a Multi-Agent Deep Reinforcement Learning (MADRL) algorithm for dynamic optimization of task offloading and resource allocation. Experimental evaluations demonstrate the proposed algorithm's superiority over traditional methods, achieving an 11% reduction in energy consumption and a 23% increase in task completion rate compared to local processing-only strategies, while reducing average task delay by 50% relative to static offloading approaches. The MADRL-based framework not only ensures efficient task distribution but also adapts to fluctuating network conditions, achieving a resource utilization rate of 85%. These findings underscore its potential to enhance performance in intelligent transportation systems by balancing computational efficiency, energy consumption, and task latency.