A dynamic task offloading method in vehicle fog environment for urban roads

Engineering notes

Key topics: autonomous driving, reinforcement learning, control. See the paper for implementation details and experimental results.

Chinese explanation / 中文解读

中文解读待补充：本站会优先为端到端自动驾驶、BEV感知、3D目标检测、轨迹预测、路径规划、LiDAR感知等高价值论文补充中文说明。

Original abstract

In urban road environments, efficient task offloading in Vehicular Ad Hoc Networks (VANETs) is critical for maximizing local resource utilization, reducing latency, and enhancing resource efficiency in dynamic vehicular fog systems. This paper proposes CDDPGM, a hierarchical task-offloading framework that combines vehicle clustering, Deep Deterministic Policy Gradient (DDPG), and inter-cluster Max-Max scheduling. The vehicle-RSU clustering stage first groups vehicles according to communication bandwidth and mobility; the DDPG scheduler then learns intra-cluster offloading decisions by jointly considering vehicle speed, bandwidth, task deadlines, and computing-resource constraints; finally, the Max-Max strategy redistributes waiting tasks among clusters when local cluster resources are insufficient. The novelty of CDDPGM lies in coupling bandwidth-aware clustering, continuous-control reinforcement learning, and cross-cluster workload redistribution in one hierarchical vehicle-fog-RSU-cloud framework, whereas most existing DRL offloading methods focus on a single RSU/MEC layer or optimize only local offloading actions. Simulation results show that, compared with NSGA-II, PSO, and Random offloading, CDDPGM reduces the average number of uncompleted tasks by 30.17%, 36.25%, and 42.23%, reduces average execution time by 2.38%, 4.65%, and 7.87%, and reduces cloud-side instruction execution by 15.34%, 18.68%, and 26.29%, respectively.

Links and sources

• PDF from original source

Comments