Multi-agent Reinforcement Learning Based Collaborative Multi-task Scheduling for Vehicular Edge Computing

Nowadays, connected vehicles equipped with advanced computing and communication capabilities are increasingly viewed as mobile computing platforms capable of offering various in-vehicle services, including but not limited to autonomous driving, collision avoidance, and parking assistance. However, providing these time-sensitive services requires the fusion of multi-task processing results from multiple sensors in connected vehicles, which poses a significant challenge to designing an effective task scheduling strategy that can minimize service requests’ completion time and reduce vehicles’ energy consumption. In this paper, a multi-agent reinforcement learning-based collaborative multi-task scheduling method is proposed to achieve a joint optimization on completion time and energy consumption. Firstly, the reinforcement learning-based scheduling method can allocate multiple tasks dynamically according to the dynamic-changing environment. Then, a cloud-edge-end collaboration scheme is designed to complete the tasks efficiently. Furthermore, the transmission power can be adjusted based on the position and mobility of vehicles to reduce energy consumption. The experimental results demonstrate that the designed task scheduling method outperforms benchmark methods in terms of comprehensive performance.