End-Edge-Cloud Collaborative Offloading of Splittable Tasks in Internet of Vehicles: A Multi-Agent Reinforcement Learning Approach with Group Updating

The rapid development of intelligent transportation and the exponential growth of data traffic drive the emerging of more computation-intensive latency-critical tasks in vehicles, bring challenges to the task offloading research in the Internet of vehicles, which provides vehicles with ultralow-latency task processing services via offloading tasks to Edge Servers (ESs) and Cloud Servers (CSs). Focusing on offloading splittable tasks, an end-edge-cloud cooperative splittable task offloading framework is presented for IoV, where vehicles with idle resources are regarded as temporary edge servers for complementing the computing services of CSs and ESs. Then, we propose a multiagent deep deterministic policy gradient algorithm to minimize task completion latency by making comprehensive decisions jointly involving task splitting, communication, and computation resource allocation. Furthermore, we propose a group-based random updating strategy for multi-agent deep reinforcement model training to promote training efficiency. To incentivize Highperformance Vehicles (HVs) to offer edge computing services via sharing their spare computation resources, we construct a Multi-Leader Multi-Follower Stackelberg incentive game model where CSs act as leaders and HVs act as followers. We prove the existence of a Stackelberg equilibrium point, and propose an optimal dynamic response algorithm to drive the CSs' resource renting price decisions and the HVs' resource sharing amount decisions to arrive at the SE point via multi-round negotiations. Simulation results demonstrate the superiority of the proposed algorithm over some selected benchmark algorithms.