6G Vehicular Metaverses

• 6G-enabled vehicular metaverses are next-gen cyber-physical platforms that fuse digital twins, ultra-low latency communications, and immersive AR/VR to enable autonomous driving and collaborative traffic control.
• Architectural frameworks leverage ISAC-equipped RSUs, vehicle sensors, and edge/cloud servers to achieve real-time digital twin updates, precise localization, and robust multi-user interactions.
• Advanced deep reinforcement learning and auction-based resource management techniques optimize VT migration and UAV routing, ensuring low latency, high semantic accuracy, and enhanced privacy.

6G-enabled vehicular metaverses integrate advanced wireless, sensing, edge/cloud, and AI technologies to deliver immersive, real-time digital environments for vehicles and users. Leveraging vehicular digital twins ("Vehicle Twins"), ultra-reliable low-latency communication, integrated sensing and communications (ISAC), semantic/task-oriented networking, and intelligent resource management, these systems extend classic Internet-of-Vehicles (IoV) architectures to full-fledged cyber-physical metaverse platforms. The resulting ecosystem supports autonomous driving, multi-user AR/VR overlays, collaborative traffic control, and personalized AI agents deployed across space–air–ground–sea integrated networks with stringent requirements for bandwidth, latency, localization precision, resource allocation, and privacy preservation.

1. Architectural Framework: ISAC and Digital Twin Integration

6G vehicular metaverse architectures center around vehicle-to-infrastructure (V2I) networks featuring roadside units (RSUs) equipped with massive MIMO arrays, in-band full duplex ISAC hardware, and waveform generators. Each vehicle (user equipment, UE) carries AR/VR displays and local sensors, interfacing with digital twin engines and multi-user synchronization servers at edge/cloud layers. The typical system configuration comprises:

• RSU with ISAC transmitter, sensing receiver, communication RF front-end
• Vehicles with communication antennas and metaverse displays
• Edge/cloud servers hosting digital twins, AR/VR rendering, physics engines
• High-capacity fronthaul/backhaul (fiber or mmWave) links

ISAC-enabled RSUs transmit OFDM frames with $N$ subcarriers and $M$ symbols at power $P$, facilitating simultaneous sensing (localization, velocity, angle) and communication (Du et al., 2023). The baseband ISAC signal model is:

$$x(t)=\sqrt{P}\sum_{m=0}^{M-1}\sum_{k=0}^{N-1} s_{m,k} e^{j 2\pi k\Delta f (t-mT)} g(t-mT)$$

Key performance metrics include throughput $R$, latency $L$, and localization accuracy $\epsilon$, with joint optimization of communication/sensing resource allocation yielding optimal rate–accuracy tradeoffs.

2. Resource Management: VT Migration, Auctions, and Air-Ground Integration

Vehicle Twin (VT) tasks, representing digital replicas of physical vehicles, require dynamic offloading to edge servers for real-time AR/AI applications. VT migration is essential for seamless service continuity as vehicles traverse RSU coverage areas. Air–ground integrated frameworks introduce UAVs as aerial edge servers to assist overloaded RSUs, alleviate resource imbalances, and maintain low latency (Tong et al., 2024, Kang et al., 2024). VT migration involves:

• Partitioning tasks into local and migrated segments for distributed processing
• Dynamic selection of RSU/UAV targets to minimize aggregate latency, maintain QoS, and avoid RSU overloads

Resource allocation in dense scenarios is complicated by information asymmetry between UAVs and ground BSs. Diffusion-based RL optimizes Modified Second-Bid (MSB) auction scaling factors, enabling adaptive, latency-aware VT task assignment with proven strategy-proofness and superior resource provider surplus compared to traditional and learning-based baselines (Kang et al., 2024).

3. Semantic and Utility-Oriented Communications

Semantic communications transcend conventional bit-oriented paradigms, instead optimizing for utility functions comprising semantic similarity, transmission delay, and power constraints. In UOC-driven vehicular metaverses, control instructions from virtual service providers (VSPs) are semantically compressed and transmitted over IRS-aided channels, where IRSs dynamically mitigate co-channel interference and improve link quality (Wang et al., 2023). The semantic utility function for $K$ users is:

$$U_{\text{sem}} = q_3\sum_{k=1}^K\xi(s_k) - q_1\sum_{k=1}^K \frac{s_k}{R_k} - q_2\sum_{k=1}^K \|\mathbf{w}_k\|^2$$

Joint optimization of semantic block lengths, beamformers, and IRS phases via alternating non-convex solvers achieves substantial gains in latency reduction, semantic accuracy maintenance ($\bar{\xi} > 0.98$), and energy efficiency compared to traditional schemes.

4. Deep Reinforcement Learning for Dynamic Resource Allocation

6G vehicular metaverse resource allocation—particularly handover management and network association under high mobility and interference—is formulated as a time-sequential, interference-coupled integer program. Deep reinforcement learning (DRL), with actor–critic architectures (e.g., A2C, PPO), effectively learns policies that minimize scene download latencies and resource waste under complex IoV mobility and channel conditions (Chua et al., 2022). State representations include data volumes, channel gains, and power allocations, while actions span cell-station associations. A2C demonstrates superior convergence and policy optimality in minimizing aggregate download times.

Diffusion-based RL architectures further enhance VT migration and UAV path planning, outperforming PPO and random migration baselines by 19–34% in asymptotic reward and 5–10% in latency reduction. DURP (dynamic UAV routing with A-Star heuristics) efficiently reduces RSU workload fluctuations and energy consumption during routing (Tong et al., 2024).

5. Integrated Sensing and Communications: Sensing-Assisted V2I Link Enhancement

ISAC advances underpin next-generation vehicular metaverses by fusing large-scale sensing (radar, localization) with high-capacity communication links. In practical 5G NR case studies, radar echo SNR at RSU receivers enables sub-ms beam steering using matched-filter and Kalman filter processing, obviating the need for CSI-RS signaling overhead. Sensing resolutions reach $\Delta R = c/(2B) \sim 3\,\text{cm}$ with $B=5\,\text{GHz}$, supporting precise digital twin updates and multi-user AR overlays under high-speed vehicular scenarios (Du et al., 2023).

6. Privacy and Security: Cross-Reality Location Protection

The proliferation of large-model AI agents and cross-reality interactions introduces non-trivial privacy challenges, notably location privacy attacks exploiting correlations between physical AV movements and virtual agent hosting. Cross-reality location entropy quantifies adversarial inference uncertainty via the Shannon entropy of the posterior given both perturbed physical location and server placement (Luo et al., 18 Jan 2026). Hybrid actions—comprising continuous location perturbations and discrete AI agent migrations—are optimized via LLM-enhanced Hybrid Diffusion Proximal Policy Optimization (LHDPPO), integrating LLM-driven reward shaping and dual generative diffusion models for tractable exploration in high-dimensional action spaces. LHDPPO achieves 15.6–27.6% utility improvement and 9.3–35% latency reduction compared to baseline privacy/migration algorithms, with robust protection against trajectory inference without hampering user immersion.

7. Open Problems and Future Directions

Several challenges remain:

• High-mobility waveforms (full-DD OTFS) to manage interference and range migration (Du et al., 2023)
• Multi-model driving cognition (IMM filtering for complex maneuvers)
• Dynamic multi-agent coordination in VT migration, multi-UAV routing, and resource auctions (Tong et al., 2024, Kang et al., 2024)
• Privacy/security protocols for cross-domain entity/service placement and data flows (Luo et al., 18 Jan 2026)
• Scalability of diffusion-based RL and auction mechanisms in ultra-dense deployments (Kang et al., 2024)
• Real-time semantic networking under rapid channel variations, IRS hardware constraints, and multi-cell orchestration (Wang et al., 2023)

This suggests that 6G-enabled vehicular metaverses will require continued integration at the intersection of wireless, sensing, AI, edge/cloud, privacy, and distributed resource management for scalable, secure, and ultra-low-latency immersive vehicle services.