Dynamic Offloading for Edge Computing-assisted Metaverse Systems

In this paper, we investigate an edge computing-assisted Metaverse system. This system involves a virtual service provider (VSP), which can partially offload sensing data collected from UAVs to an edge computing platform. The data is used to update its digital twins (DTs) to ensure the promptness of Metaverse services and satisfy the latency requirements of Metaverse users. However, designing such a system is challenging due to the dynamics of sensing data, the latency requirements of Metaverse users, channel conditions, and the available computing resources at both the VSP and EC. Therefore, we formulate the VSP's offloading problem as a stochastic problem and utilize deep reinforcement learning (DRL) algorithms. Simulation results are provided to validate the effectiveness of the learning algorithms.