A Cybertwin-Driven Intelligent Offloading Method for IoV Applications Using DRL in Smart Cities

Internet of Vehicle (IoV) is considered as the enabling platform to improve the traffic condition in cities. Meanwhile, a variety of promising applications such as self-driving and route planning are realized, which generates various latency-critic tasks. These latency-critic tasks present high demands on computing capacity, while resource-restricted vehicles could not satisfy those requirements. Mobile edge computing (MEC) is emerging as a potential solution to mitigate resource-limited vehicles from latency-critic tasks, which enables vehicles to offload tasks to nearby road-side units. Besides, cybertwin can abstract IoV system into a digital entity to facilitate obtaining information and management. However, it is still challenging to leverage cybertwin technology to optimize offloading policy. To this end, we design a deep reinforcement learning method to train the optimal decision-making agent by a cybertwin-driven IoV system, which aims to minimize the system cost under latency constraints. The experimental results illustrate that the agent trained by our proposed method outperforms the other methods.