PaperlayerView exact revision on arXiv

Source-linked AI summary

A Full Dive into Realizing the Edge-enabled Metaverse: Visions, Enabling Technologies,and Challenges

Minrui Xu, Wei Chong Ng, Wei Yang Bryan Lim, Jiawen Kang, Zehui Xiong, Dusit Niyato, Qiang Yang, Xuemin Sherman Shen, Chunyan Miao

arXiv:2203.05471v2cs.NIcs.CY

TL;DR

Existing Metaverse implementations remain limited by interoperability and by the communication and computation demands of immersive access, particularly on resource-constrained edge devices. This survey examines edge-enabled architectures, networking and computation solutions, AI techniques, blockchain roles, and future research directions. The surveyed approaches include federated viewpoint prediction, FoV-aware caching, light-field streaming, cloud-edge-end computation, knowledge distillation, and adversarial defenses.

  • Problem

    Current Metaverse implementations are not yet interoperable or capable of supporting immersive, resource-intensive access for billions of users, especially on constrained edge devices.

  • Method

    The survey synthesizes communication and networking, cloud-edge-end computation, AI, blockchain, and future research directions for the edge-enabled Metaverse.

  • Results

    The survey presents edge-oriented solutions including federated viewpoint prediction, FoV-aware caching, light-field streaming, distributed computation, knowledge distillation, and adversarial defenses.

  • Takeaways & Limitations

    Realizing the Metaverse requires coordinated communication, computation, and blockchain infrastructure that supports immersive, interoperable access at the network edge.

Abstract

from arXiv · show

Dubbed "the successor to the mobile Internet", the concept of the Metaverse has grown in popularity. While there exist lite versions of the Metaverse today, they are still far from realizing the full vision of an immersive, embodied, and interoperable Metaverse. Without addressing the issues of implementation from the communication and networking, as well as computation perspectives, the Metaverse is difficult to succeed the Internet, especially in terms of its accessibility to billions of users today. In this survey, we focus on the edge-enabled Metaverse to realize its ultimate vision. We first provide readers with a succinct tutorial of the Metaverse, an introduction to the architecture, as well as current developments. To enable ubiquitous, seamless, and embodied access to the Metaverse, we discuss the communication and networking challenges and survey cutting-edge solutions and concepts that leverage next-generation communication systems for users to immerse as and interact with embodied avatars in the Metaverse. Moreover, given the high computation costs required, e.g., to render 3D virtual worlds and run data-hungry artificial intelligence-driven avatars, we discuss the computation challenges and cloud-edge-end computation framework-driven solutions to realize the Metaverse on resource-constrained edge devices. Next, we explore how blockchain technologies can aid in the interoperable development of the Metaverse, not just in terms of empowering the economic circulation of virtual user-generated content but also to manage physical edge resources in a decentralized, transparent, and immutable manner. Finally, we discuss the future research directions towards realizing the true vision of the edge-enabled Metaverse.

I. INTRODUCTION

The Metaverse is envisioned as an immersive, embodied, and interoperable successor to today’s Internet, but current implementations remain limited. This survey therefore examines the communication, computation, and blockchain challenges of realizing it at mobile edge networks.

  • The Metaverse is commonly described as an embodied Internet navigated through AR, VR, and tactile interfaces.
  • Current “lite” implementations remain separate virtual worlds, whereas the full vision requires seamless interoperability and portable avatar belongings and assets.
  • Realizing the Metaverse requires implementation solutions for communication, networking, computation, and resource-constrained edge devices.
  • The survey reviews communication and networking challenges, including massive user interactions and differentiated service requirements, rather than only conventional AR/VR applications.
  • It surveys cloud-edge-end computation and AI techniques for improving Metaverse access and computational efficiency at the network edge.
  • It also examines blockchain’s economic and infrastructure roles and outlines future research directions for mobile edge networks.

II. ARCHITECTURE, DEVELOPMENTS, AND TOOLS OF THE METAVERSE

The Metaverse architecture links physical and virtual ecosystems through immersive interaction, synchronization, and interoperable services. Its realization depends on communication, computation, and blockchain infrastructure, including edge-based support for demanding applications.

  • The Metaverse comprises interoperable, immersive, and shared virtual ecosystems navigable by user avatars.
  • Physical-virtual synchronization: Physical-virtual synchronization lets stakeholder actions affect virtual worlds and enables feedback from virtual environments to the physical world.
  • Physical-virtual synchronization: IoT and sensor networks collect physical-world data that updates virtual worlds, including digital twins maintained from live sensing feeds.
  • Enabling technologies: AR/VR and haptics provide visual and tactile interaction, while the tactile Internet targets approximately 1 ms round-trip delay for haptic and kinesthetic information.
  • Computation: Cloud-edge-end computation distributes avatar, rendering, and data-processing tasks across devices, edge servers, and the cloud to reduce latency for mobile users.
  • Blockchain: Blockchain supports proof of ownership for virtual goods and reduces reliance on centralized platforms through decentralized records.

B. Current Developments

The Metaverse remains in an early “proto-verse” stage, with development progressing across physical and virtual worlds. Physical-virtual synchronization is presented as a route toward linking digital and physical experiences.

  • The Metaverse is still in its infancy, described as a “proto-verse” requiring substantial further development.
  • Its development depends on communication and networking, computation, and blockchain infrastructure supporting physical-to-virtual and virtual-to-physical synchronization.
  • A real-time virtual 3D environment can reflect the physical world to support remote working, socializing, and services such as education.
  • Representative developments include virtual cities, theme-park initiatives, MMORPG-like worlds, user-created environments, and blockchain-based games.

2) Applications of the Metaverse:

Metaverse applications span government, tourism, education, healthcare, entertainment, and industrial collaboration, supported by platforms and development tools. Examples range from avatar-based services to virtual replicas and shared events.

  • Metaverse initiatives include smart cities, entertainment, education, working environments, and healthcare applications.
  • Metaverse Seoul combines a virtual city hall, tourism destinations, social services, avatar officials, and virtual festivals.
  • Xirang supports smartphone and VR access, automatically generated avatars, and a virtual AI developers’ conference for 100,000 users at once.
  • CUHKSZ provides a blockchain-powered virtual campus where real-world behaviors can affect the virtual world and vice versa.
  • Telemedicine applications let patients and healthcare professionals meet virtually while physical tests and scans remain locally performed.
  • Development tools and platforms include Unity, Unreal Engine, Roblox, Nvidia Omniverse, Meta Avatars, HoloLens2, and Oculus Quest2.

D. Lessons Learned

The Metaverse emerges from converging technologies that synchronize physical and virtual worlds, rather than from any standalone technology. Its development also faces interoperability, accessibility, and user-contribution challenges.

  • D. Lessons Learned: The Metaverse combines AR/VR, tactile Internet, digital twins, AI, and blockchain-based economy for physical-virtual synchronization.Each technology contributes a distinct capability, but none alone constitutes the Metaverse.
  • D. Lessons Learned: AR/VR provides 3D virtual and auditory content, while the tactile Internet supplements it with tactile and kinesthetic content.
  • D. Lessons Learned: Digital twins connect physical and virtual entities and support synchronization, but do not directly provide immersive user experiences.
  • D. Lessons Learned: AI improves physical-virtual synchronization efficiency but does not directly facilitate synchronization.
  • D. Lessons Learned: Current Metaverse projects require computation-intensive rendering, real-time digital-twin updates, and low-latency reliable communications.
  • D. Lessons Learned: Standardized protocols, frameworks, and tools would make content transferable across virtual worlds and support smoother user transitions.
  • D. Lessons Learned: User-generated content and open-source contributions support the Metaverse ecosystem, while digital currencies can facilitate exchanges and payments.

III. COMMUNICATION AND NETWORKING

Communication and networking must provide accessible, immersive, and real-time Metaverse experiences despite demanding traffic and limited edge resources. The survey reviews streaming, prediction, caching, multimodal delivery, and economic mechanisms for these requirements.

  • III. COMMUNICATION AND NETWORKING: Mobile edge networks must provide high-speed, low-delay, wide-area wireless access for seamless and real-time Metaverse experiences.
  • III. COMMUNICATION AND NETWORKING: Metaverse networking must support immersive 3D streaming, multisensory communication, and stringent real-time interaction requirements.Relevant constraints include motion-to-photon, interaction, and haptic perception latency.
  • III. COMMUNICATION AND NETWORKING: The survey reviews seamless 3D multimedia delivery, user-centric content delivery, semantic or goal-oriented communication, and digital-twin synchronization.
  • III. COMMUNICATION AND NETWORKING: AR/VR networking must jointly optimize data rate, interaction latency, and reliability, whose requirements can vary dynamically across users and applications.
  • Resource Allocation for VR Streaming: VR streaming faces resource-allocation challenges because edge-network spectrum and wireless-device energy resources are limited.
  • Resource Allocation for VR Streaming: VR delivery studies consider heterogeneous transmission modes, including macrocell broadcasting, small-cell unicasting, and device-to-device multicast.
  • Resource Allocation for VR Streaming: Federated viewpoint prediction with a VR stochastic buffering game and Dueling Double Deep Recurrent Q-Network improves viewing utility by 60% over the baseline algorithm.
  • Resource Allocation for VR Streaming: Multimodal VR streaming must reconcile visual content’s high-rate needs with haptic content’s fixed-rate and very-high-reliability requirements.

2) Resource Allocation for AR Adaptation:

Edge-enabled Metaverse access requires adapting AR services, coordinating shared content, and delivering immersive media under constrained computation, communication, and latency resources. Reviewed approaches trade accuracy, quality, privacy, bandwidth, and reliability against these constraints.

  • 2) Resource Allocation for AR Adaptation:: Mobile edge networks support precise AR and approximate AR, with approximate AR reducing edge computation and communication overhead by detecting only attended targets.
  • 2) Resource Allocation for AR Adaptation:: Higher AR service latency generally supports higher analysis accuracy, motivating network coordination to balance accuracy against latency.
  • 2) Resource Allocation for AR Adaptation:: Multi-user collaborative MAR frameworks address cross-platform interaction, efficient communication, and heterogeneous users through prediction-based key-frame selection.
  • 3) Edge Caching for AR/VR Content:: Distributed edge caching must reduce transmission overhead while preserving uniform Metaverse content across geographically distributed edge servers.
  • 3) Edge Caching for AR/VR Content:: FoV-aware panoramic VR caching improves cache hit ratio and bandwidth saving by a factor of two, reducing network latency and bandwidth pressure.
  • 4) Hologram Streaming:: Wireless holographic communication can project detailed information about users into the physical world, using optical or computer-generated holography and digital holography.
  • 4) Hologram Streaming:: Point-cloud streaming adapts immersive 3D content to device capabilities and available communication and computation resources to reduce unnecessary transmission.
  • 4) Hologram Streaming:: Light-field streaming supports glasses-free, multi-angle interaction; personalized view selection and encoding can reduce required bitrate by around 50%.

1) URLLC for the tactile Internet:

The tactile Internet supplements AR/VR with haptic and kinesthetic interaction, imposing approximately 1 ms end-to-end latency and stringent reliability requirements. Semantic communication addresses Metaverse traffic by transmitting task-relevant meaning rather than reconstructing source messages.

  • 1) URLLC for the tactile Internet:: The Metaverse requires haptic and kinematic interaction in addition to 360° visual and auditory content.The tactile Internet supports this embodied interaction alongside AR/VR.
  • 1) URLLC for the tactile Internet:: Approximately 1 ms of end-to-end latency is required for tactile information exchange during teleoperation.Tactile interaction is more sensitive to system stability and latency than visual and auditory information.
  • 1) URLLC for the tactile Internet:: URLLC is required for human-human and human-avatar tactile interaction, while AR/VR primarily relies on eMBB.Bursty tactile-packet arrivals further complicate wireless-edge resource allocation.
  • 1) URLLC for the tactile Internet:: Semantic communication transmits only the information required for a task or goal instead of restoring the original source message.Its semantic encoder removes irrelevant information, while the decoder restores an understandable representation or corresponding action.
  • 1) URLLC for the tactile Internet:: Semantic communication evaluates errors using human-perception measures such as sentence similarity rather than conventional bit error rate.Multimodal extensions and lightweight neural networks target resource-constrained devices.

1) Resource Allocation for Physical-Virtual Synchronization:

Physical-virtual synchronization requires efficient resource allocation across sensing, communication, computation, and wireless-environment control. Surveyed mechanisms use incentives, distributed optimization, digital twins, RIS, and integrated networks to improve efficiency and coverage.

  • 1) Resource Allocation for Physical-Virtual Synchronization:: Incentive mechanisms address self-interested edge devices and vehicles contributing communication, sensing, and energy resources to physical-virtual synchronization.A distributed alternating-direction method of multipliers mechanism maximizes energy efficiency in dynamic vehicular twin networks.
  • 1) Resource Allocation for Physical-Virtual Synchronization:: Resource allocation in digital-twin networks jointly considers offloading, transmission power, bandwidth, and computation to balance resource consumption and service delay.The formulation accounts for probabilistic synchronization-task arrivals and complex network topology.
  • 1) Resource Allocation for Physical-Virtual Synchronization:: Digital twins can model edge-network topology and channels, enabling communication and coordination that improve real-world edge-network performance.They support high-fidelity modeling, simulation, and prediction of physical entities.
  • 1) Resource Allocation for Physical-Virtual Synchronization:: RIS can adaptively phase its elements to modify the wireless environment, but jointly optimizing RIS phases and beamforming power remains challenging.RIS can also support immersive-content streaming.
  • 1) Resource Allocation for Physical-Virtual Synchronization:: IoV, UAV, RIS, and satellites can form a SAGIN to provide ubiquitous Metaverse coverage and support secure vehicle communication.A virtual-space digital-twin network enables physical entities to communicate through virtual entities.
  • 1) Resource Allocation for Physical-Virtual Synchronization:: Context-aware caching uses users’ FoVs, view angles, and interacting objects to reduce perceived latency and bandwidth for immersive streaming.The approach places immersive content at edge providers based on contextual preferences.

IV. COMPUTATION

Metaverse computation spans high-dimensional data processing, 3D rendering, and avatar intelligence, but edge devices are resource constrained. The survey presents cloud-edge-end collaboration as a way to place computation near data sources and support ubiquitous, low-latency access.

  • IV. COMPUTATION: Metaverse applications require high-dimensional data processing, 3D virtual-world rendering, and avatar computing.These workloads support realistic worlds, displayable 3D objects, and intelligent real-time avatar interaction.
  • IV. COMPUTATION: Resource-constrained user devices need ubiquitous computation and intelligence for offloading, embodied telepresence, personalized recommendations, and cognitive avatars.User mobility and stochastic demand require adaptive and continuous edge services.
  • IV. COMPUTATION: Cloud-edge-end collaboration distributes computation across user devices, edge infrastructure, and cloud resources to provide ubiquitous Metaverse computing.The survey organizes solutions around cloud-edge-end rendering and AI-driven techniques.
  • IV. COMPUTATION: Edge computing provides low latency, high efficiency, and security by using computation and storage closer to users than the cloud.It also reduces backbone-network burden by processing tasks in the local network.
  • IV. COMPUTATION: Fog computing extends cloud services closer to users because distant-cloud transmission latency cannot support real-time immersive interaction.The cited theoretical results show lower latency than traditional cloud computing.
  • IV. COMPUTATION: Cloud-edge-end architectures execute computation where data is generated, reducing end-to-end latency and shifting traffic from the Internet to the edge.Nearby vehicles or roadside units can process interaction data instead of relying on cloud offloading.

1) Stochastic Demand and Network Condition:

Efficient Metaverse rendering must adapt to changing demand, network conditions, heterogeneous tasks, and stragglers. The surveyed approaches combine local or edge offloading, coded distributed computing, adaptive learning, and model compression to reduce latency and resource demands.

  • 1) Stochastic Demand and Network Condition:: AR devices can compute locally under poor network conditions or offload rendering tasks to edge servers when network conditions permit.Adaptive offloading reduces dependence on individual devices and addresses stragglers affecting shared immersive experiences.
  • 1) Stochastic Demand and Network Condition:: Polynomial codes accelerate distributed matrix multiplication and achieve a recovery threshold that does not increase with the number of worker nodes.Vehicular collaborative computing combines coded distributed computing with reputation and game-theoretic worker selection.
  • 1) Stochastic Demand and Network Condition:: MatDot and PolyDot codes support smooth performance degradation and low-complexity decoding under fixed deadlines in edge computing.They address the encoding and decoding costs created by many physical entities.
  • 1) Stochastic Demand and Network Condition:: The SIP offloading scheme has lower cost than a fixed-straggler scheme because it models stragglers’ shortfall uncertainty.Its objective is to minimize UAV energy consumption while mitigating stragglers.
  • 1) Stochastic Demand and Network Condition:: Metaverse AR/VR rendering policies must accommodate heterogeneous tasks with different delay requirements, such as stricter foreground than background rendering latency.Meta reinforcement learning can adapt across environments using limited gradient updates and samples.
  • 1) Stochastic Demand and Network Condition:: Pruning, quantization, and Huffman coding compress deep-learning models without losing accuracy, reducing AlexNet storage by 35 times and VGG-16 storage by 49 times.The resulting lightweight models can be stored on edge devices for intelligent Metaverse applications.

2) Low-rank Approximation:

Low-rank approximation compresses large AI models by factorizing matrices into smaller ones, reducing edge-device memory requirements while supporting Metaverse deployment. Combined with quantization, it can preserve or slightly improve test accuracy.

  • 2) Low-rank Approximation:: Low-rank approximation factorizes large matrices into smaller matrices, compressing AI layers and reducing the model memory footprint on edge devices.It complements pruning, which removes redundant parameters.
  • 2) Low-rank Approximation:: One-shot whole network compression uses rank selection, Tucker decomposition, and fine-tuning to deploy deep-learning models on edge devices.Variational Bayesian matrix factorization determines ranks for kernel tensors before decomposition and accuracy recovery.
  • 2) Low-rank Approximation:: Combining low-rank approximation with quantization can produce slightly higher test accuracy than the corresponding non-compressed models.The two methods exploit different redundancy sources: over-parameterization and parameter representation.
  • 3) Knowledge Distillation:: Knowledge distillation transfers a large teacher model’s knowledge to a smaller student model for edge deployment.The teacher provides predicted class probabilities as soft targets, controlled by temperature T.
  • 3) Knowledge Distillation:: FitNets use a deeper, thinner student that reduces computational burden while improving generalization relative to the teacher.In the smallest network, the student is 36 times smaller with a 1.3% performance drop.

2) Federated Learning:

Federated learning trains Metaverse models across edge devices without centralizing user data. It reduces communication costs and supports continual, private, and personalized learning, but remains exposed to poisoning, inference, and privacy-performance trade-offs.

  • 2) Federated Learning:: Federated learning trains models across edge devices without moving application data to a centralized location.Unlike distributed learning, FL transmits model parameters rather than first collecting and redistributing raw data.
  • 2) Federated Learning:: FL reduces communication costs, supports continual learning from fresh sensor data, protects privacy, and enables personalized services.Only model parameters are transmitted, while local training uses users’ latest data.
  • 2) Federated Learning:: FL still has data-privacy limitations despite keeping personal data on edge devices.The remaining risks include information leakage through model updates and recovery of users’ training samples.
  • 2) Federated Learning:: Distributed FL increases attack surfaces because malicious users can inject mislabeled or manipulated data into local training.Such data poisoning can mislead the global model.
  • 2) Federated Learning:: Differential-privacy techniques can significantly reduce model accuracy when few edge devices participate in aggregation.With more participating devices, DP and non-DP model accuracies become similar.
  • 3) Adversarial Machine Learning:: MULDEF improves robustness against adversarial examples by maintaining multiple models and randomly selecting one at runtime.More models increase the chance of selecting a robust model.

E. Lessons Learned

Edge-enabled Metaverse systems combine collaborative computation, model compression, privacy protection, and blockchain-based decentralization. The survey identifies interoperability, distributed storage, and single-point-of-failure concerns as continuing boundaries.

  • E. Lessons Learned: Cloud-edge-end collaboration offloads AR/VR rendering to available computing nodes and adaptively addresses demand, network, straggler, and task heterogeneity.Coordination across ubiquitous computing and intelligence resources targets performance bottlenecks in interactive rendering.
  • E. Lessons Learned: Model compression reduces local hardware requirements, but on-demand compression must balance model accuracy against users’ quality of experience.Resource constraints include storage, computational power, and energy.
  • E. Lessons Learned: Metaverse systems require trustworthy, privacy-preserving handling of sensitive sensor information, including GPS locations, voice, and eye movements.Protection is needed across both virtual applications and physical-world data captured by AR/VR devices.
  • E. Lessons Learned: Separate edge or cloud storage and differing virtual-world development frameworks complicate offloading across physical and virtual worlds.Distributed storage is identified as important for traversal between these worlds.
  • E. Lessons Learned: Cloud-edge and federated computation can involve centralized servers, creating a single point of failure that decentralized storage and offloading aim to address.The survey contrasts distributed computation with later blockchain-based decentralization.
  • E. Lessons Learned: Blockchain can record virtual content, avatar data, and edge resources to support ownership, secure governance, interoperability, and automatic resource management.Smart contracts and consensus algorithms provide decentralized, tamper-proof coordination without third-party authorities.

1) User-generated Content:

User-generated content lets Metaverse users create virtual assets, while blockchain and decentralized storage support ownership, monetization, and data availability. However, storage risks and limited interoperability remain unresolved.

  • Users can create virtual assets such as videos, music, and profile pictures instead of relying solely on platform-provided marketplaces.
  • Blockchain-based UGC frameworks can place heterogeneous data and edge resources near data sources, reducing service delay and network load.
  • NFTs tokenize user-generated content by recording unique identifiers and encrypted transaction histories that verify ownership and support future valuation.
  • Centralized media storage can expose NFT systems to data loss and denial-of-service risks, while existing NFT schemes remain limited by blockchain interoperability.
  • 3) Game and Social Financial Systems:: GameFi and SocialFi combine edge or cloud access, blockchain-based trading, and player-created asset value within Metaverse economic systems.
  • Off-chain systems such as IPFS and Storj provide decentralized, redundant, or encrypted storage for large Metaverse datasets.

C. Blockchain Scalability and Interoperability

Blockchain scalability and interoperability are needed to connect multiple physical and virtual entities, support cross-chain value and data exchange, and coordinate edge services. Surveyed approaches include sharding, cross-chain frameworks, multi-chain slicing, and blockchain-assisted edge security.

  • C. Blockchain Scalability and Interoperability: Cross-chain technology connects blockchain networks to exchange information and value, addressing interoperability across multiple Metaverse virtual worlds.
  • C. Blockchain Scalability and Interoperability: Network sharding divides edge nodes into local blockchain shards while cloud nodes manage a global blockchain to improve access-control scalability.
  • C. Blockchain Scalability and Interoperability: Cross-chain frameworks can integrate multiple blockchains through notary mechanisms for secure and efficient IoT data management.
  • C. Blockchain Scalability and Interoperability: Multi-chain network slicing deploys a separate blockchain per edge slice and uses a service-quality chain to coordinate contracts across slices.
  • D. Blockchain in Edge Resource Management: Blockchain supports Metaverse communication and computation through access control, auditing, transparent task processing, and privacy-preserving protection of AI models.
  • D. Blockchain in Edge Resource Management: Wireless blockchain scenarios include NFT-based virtual-content trading and interoperation among blockchains for economic systems, avatar management, and edge resources.

3) Blockchain in Edge Computation Offloading:

Blockchain can support computation offloading, resource coordination, data management, and decentralized edge intelligence in resource-constrained Metaverse environments. The surveyed approaches also address trust, privacy, storage limits, and physical-virtual economic synchronization.

  • 3) Blockchain in Edge Computation Offloading:: Blockchain-enabled offloading and resource-allocation frameworks reduce computation loads for VR devices performing edge medical-treatment services.
  • 3) Blockchain in Edge Computation Offloading:: Blockchain-based collaborative edge platforms use layered trust structures to motivate selfish edge servers and accommodate heterogeneous vehicular-cloud resources.
  • 3) Blockchain in Edge Computation Offloading:: Blockchain provides decentralized management and sharing of data and knowledge resources among edge devices, but efficiency and risk worsen as user-data volume and privacy demands increase.
  • 3) Blockchain in Edge Computation Offloading:: Federated learning can collaboratively train global models for avatars and recommendations, although unreliable local models can degrade global-model performance through data poisoning.
  • 3) Blockchain in Edge Computation Offloading:: Physical and virtual monetary systems can synchronize through P2V access, digital-asset trading, and V2P conversion to fiat currencies.
  • 3) Blockchain in Edge Computation Offloading:: Off-chain storage keeps complete data off-chain while verifying and storing metadata on-chain, reducing storage pressure while preserving immutability.
  • 3) Blockchain in Edge Computation Offloading:: AI can support adaptive blockchain sharding and learning-based notary selection, while blockchain can decentralize edge training and inference and improve network reliability.

VI. FUTURE RESEARCH DIRECTIONS

Future research targets embodied user experience, sustainable edge networks, and extensive edge intelligence for the Metaverse. Priorities include immersive communication, physical-virtual synchronization, resource-efficient digital twins, intelligent blockchain, QoE, and service-market design.

  • Advanced Multiple Access for Immersive Streaming: Advanced multiple access must support ubiquitous accessibility, extremely high data rates, and ultra-reliable low-latency communications for heterogeneous immersive services.
  • Multi-sensory Multimedia Networks: Multi-sensory multimedia networks require resource allocation across simultaneous AR/VR, tactile Internet, and hologram services with different requirements.
  • Multimodal Semantic/Goal-aware Communication: Multimodal semantic communication must support multi-sensory services while efficiently extracting semantics and allocating edge resources for model training and application.
  • Integrated Sensing and Communication: Integrated sensing and communication can support real-time digital replication of physical entities and distribute synchronized information to users.
  • Digital Edge Twin Networks: Digital edge twins can synchronize, monitor, and control physical entities, but their communication, computation, and storage demands challenge resource-limited edge networks.
  • Edge Intelligence and Intelligent Edge: Edge intelligence combines Edge for AI, which brings the sensing-to-inference pipeline closer to data, with AI for Edge, which improves its orchestration.
  • Future work also includes privacy-aware avatar services, intelligent blockchain linked to physical entities, QoE assessment, and markets coordinating service allocation and pricing.
  • The survey frames future directions around embodied experience, sustainable edge networks, and extensive edge intelligence supported by physical-virtual synchronization and blockchain.
Loading 2203.05471v2…