Multi-user QoE optimization in virtual reality gaming using RL on mobile edge networks

Abstract In this paper, we explore delivering mobile edge Virtual Reality (VR) gaming services with comprehensive and satisfactory Quality of Experience (QoE) across a distributed edge network. Our goal is to meet the QoE needs of all users, addressing both latency and visual considerations. We systematically capture the unique attributes of mobile VR games, including numerous independent objects with distinct rendering pipelines, diverse rendering levels, tight end-to-end delay requirements, and high bandwidth usage. We demonstrate that this challenge can be formulated as a Mixed-Integer Quadratically Constrained Quadratic Programming (MIQCQP) problem. By leveraging reinforcement learning techniques, we introduce a flexible and adaptive QoE weight formulation that incorporates symbolic generalization and real-time feedback loops. Unlike existing studies that often focus on single-user scenarios, our paper comprehensively addresses visual and latency concerns across multiple users. We employ optimization techniques through eigenvalue analysis and Lagrange-constrained methods, providing a scalable and robust framework for dynamically adjusting QoE weights based on user interactions and environmental factors. This approach enables us to deliver QoE-centric edge-assisted mobile VR gaming services to a large user base, advancing beyond current limitations in the field by leveraging reinforcement learning (RL), specifically Q-Learning, to dynamically adapt to the non-linear, real-time demands of multi-user VR gaming on edge networks.