Joint Task Offloading and Resource Allocation in Vehicular Platoon Networks: A Federated R...

In the future, NR-V2X networks and platoon-based driving modes will become essential components of Intelligent Transportation Systems. However, with the advent of the Internet of Vehicles, the large-scale deployment of sensors and the explosive growth of data are driving the need for new solutions. Mobile edge computing has emerged as a key technology for addressing the distributed computing requirements. In this paper, we propose a Joint Optimization framework based on Federated multi-agent Reinforcement Learning (JOFRL) to solve the task offloading and resource allocation problems in a platoon-based NR-V2X network. Existing RL algorithms focus on either the communication link capacity or the completion rate of computing tasks. Differently, we consider the mutual influence between the allocation of computing and communication resources. We modify the reward function in the MARL framework so that each agent’s communication performance on V2V links is aligned with its computing performance. Our experimental results demonstrate that JOFRL outperforms other baseline algorithms. Specifically, JOFRL achieves improvements of 10.87%, 22.43% and 23.02% in computing and communication respectively compared with MADDPG, SAC and DDPG algorithm.

Joint Task Offloading and Resource Allocation in Vehicular Platoon Networks: A Federated Reinforcement Learning Approach

Taomin Wang, Yiming Liu, Qiang Wang, Xuguang Cao, Jingyi Chen, Beijing University of Posts and Telecommunications