Deep Reinforcement Learning for the Optimization of Metaverse Service Quality

The Metaverse aims to provide immersive experiences as the next-generation communication paradigm. This requires balancing high-quality visuals, low latency, and head-mounted device (HMD) energy consumption. However, developing a unified framework that jointly models these factors and optimizes their trade-offs in real-time remains challenging. We construct a comprehensive mathematical model for Metaverse user utility, centering on video resolution while treating latency and energy consumption as penalty terms. We then design a deep reinforcement learning method based on Proximal Policy Optimization (PPO) that dynamically optimizes computational task allocation according to available resources on local devices and edge servers. Experimental results demonstrate our approach’s effectiveness: our method achieves 38% improvement in user utility compared to the state-of-the-art Meta-Immersion model, with notable advantages in high-resolution services. Additionally, our PPO-based algorithm minimizes user cost more effectively than baseline strategies by intelligently utilizing edge resources. This research delivers both an accurate user utility model and an efficient optimization algorithm for resource-constrained Metaverse environments.