Multi-Agent Deep Reinforcement Learning Based on Actor-Critic for Task Offloading

5G-enabled Vehicle Edge Computing (VEC) offers a promising approach, allowing computationally demanding jobs to be transferred from smart cars to edge Internet of vehicles (IoV) devices. Numerous applications are made possible by this, such as autonomous driving and on-board AR/VR. However, a primary obstacle facing VEC is the imperative to minimize latency and energy consumption during task offloading. This study introduces a novel VEC task offloading framework leveraging multi-agent deep reinforcement learning (DRL), we aim to optimize the multi-vehicle-assisted Mobile Edge Computing (MEC) system, recognizing the constraints posed by vehicle computing resources on timely task completion. This problem can be solved by assigning jobs to roadside units (RSUs) with more powerful computers. Using a DRL method that is Actor-Critic improves system performance and convergence efficiency. The advantage of our proposed DRL method over the deep Q-learning strategy is demonstrated by the simulation results. significantly accelerating convergence speed, enhancing performance, and reducing overall system costs.