Research on UAV-assisted Vehicle networking task unloading strategy based on multi-agent reinforcement learning

Abstract

With the development of technology, in order to improve the user’s driving experience and driving safety, there are more and more vehicle tasks with high delay requirements.  Therefore, lots of researchers have paid attention to task offloading scheduling.However, as vehicle tasks become increasingly complex, a single task may consist of multiple subtasks with dependencies between them.The complex data dependencies within them make it more and more difficult to design appropriate task offloading strategies. Considering that this problem is closely related to the scenarios and requirements in the real world, this study focuses on the design of task offloading decisions in the scenario of UAV-assisted vehicle network, in which MEC servers are installed in the macro base station and UAV to provide computing resources for vehicles. We designed a task offloading strategy based on MATD3 algorithm to deal with this problem. Following simulation trials, it is evident that our approach offers notable benefits in terms of both delay and energy usage.