Abstract
The Metaverse can be considered the extension of the present-day web, which integrates the physical and virtual worlds, delivering hyper-realistic user experiences. The inception of the Metaverse brings forth many ecosystem services such as content creation, social entertainment, in-world value transfer, intelligent traffic, and healthcare. These services are compute-intensive and require computation offloading onto a Metaverse edge computing server (MECS). Existing Metaverse edge computing approaches do not efficiently and effectively handle resource allocation to ensure a fluid, seamless, and hyper-realistic Metaverse experience. Therefore, we introduce a new Metaverse-compatible, Unified, User and Task (UUT) centered artificial intelligence (AI)-based mobile edge computing (MEC) paradigm, which serves as a concept upon which future AI control algorithms could be built to develop a more user and task-focused MEC.
Introduction
The Metaverse promises a seamless blend of physical and digital realities, offering "hyper-realistic" experiences such as 8K 360-degree video and sub-millisecond haptic feedback. However, these applications present significant computational challenges. To support lightweight user devices like VR glasses, heavy computations must be offloaded to nearby servers via Mobile Edge Computing (MEC).
Current MEC resource allocation strategies face three primary hurdles. First, extreme Quality of Service (QoS) demands require high frame rates (120 FPS) and ultra-low latency (<20ms) to prevent motion sickness. Second, user heterogeneity means devices vary wildly in capabilities and battery states. Third, task complexity involves managing simultaneous sub-objectives like rendering, movement prediction, and semantic data extraction.
Methodology: The UUT-Centered AI Framework
To address these challenges, we propose a Unified, User and Task (UUT) centered Deep Reinforcement Learning (DRL) architecture. This framework shifts away from traditional "one-size-fits-all" monolithic AI models.
User-Centered DRL
Instead of a single "Critic" evaluating the entire network state, the user-centered approach employs a multi-tail architecture. The Critic generates individual state values ($V_i$) and advantage values ($A_i$) for each user. This allows the AI to make granular observations about individual user experiences and adjust resources dynamically based on specific device constraints and user needs.
Task-Centered DRL
The task-centered component utilizes a Multi-Actor setup where different agents handle specific tasks, such as channel access or resolution allocation. A Hybrid Critic evaluates these tasks against both their specific sub-objectives and a global objective. This ensures that individual tasks do not "hog" resources at the expense of the overall system performance, solving the "selfish agent" problem common in multi-agent systems.
Semantic Communication and Predictive Modeling
To optimize bandwidth, the framework incorporates semantic communication using AI autoencoders to extract and transmit essential features rather than raw data. Furthermore, the system treats Metaverse physics as a data-driven service, using DRL to predict user movement (tracking vectors) and pre-allocate resources to where the user is likely to look next.
Experimental Evaluation and Results
The UUT framework was evaluated using two specialized simulation environments mimicking Metaverse workloads: a VR Scene Offloading simulation and a Smart Vehicle Digital Twinning scenario.
In VR rendering tests, the UUT-centered RL achieved a significantly higher task success rate (successful renders per second) compared to traditional RL, which struggled to converge on an efficient policy. In the smart vehicle case study, which balanced object detection accuracy (mAP) against uplink latency, the proposed model achieved the lowest total delay (~10ms) while maintaining the highest mAP. Traditional separate RL agents were nearly twice as slow and significantly less accurate.
Conclusion and Future Work
The UUT-centered AI paradigm offers a robust foundation for managing the complex resource demands of the Metaverse. By focusing on individual user needs and specific task requirements, the framework achieves superior performance in latency and rendering success.
Future research will focus on addressing privacy concerns inherent in user-centered data collection through techniques like Privacy-Preserving Edge Computation. Additionally, we aim to better model subjective Quality of Experience (QoE) and integrate centralized cloud coordination for long-term, background world-building tasks.
Summary updated