Cross-reality Location Privacy Protection in 6G-enabled Vehicular Metaverses: An LLM-enhanced Hybrid Generative Diffusion Model-based Approach

TL;DR

This work tackles cross-reality location privacy for agentic AI-driven AVs in 6G vehicular metaverses, where adversaries can fuse real LBS locations with virtual agent hosting. It introduces a hybrid action framework combining continuous location perturbation in reality with discrete AI agent migration in virtuality, and a new cross-reality location entropy metric to quantify privacy. An LLM-enhanced Hybrid Diffusion Proximal Policy Optimization (LHDPPO) algorithm is developed to solve the resulting non-convex mixed-integer problem, utilizing an LLM-driven reward design loop and dual diffusion-based policies for continuous and discrete actions. Experiments on real mobility and network data show improved privacy (higher $E_m^t$) and lower latency while maintaining immersion, illustrating practical potential for scalable cross-reality privacy protection in 6G vehicular metaverses.

Abstract

The emergence of 6G-enabled vehicular metaverses enables Autonomous Vehicles (AVs) to operate across physical and virtual spaces through space-air-ground-sea integrated networks. The AVs can deploy AI agents powered by large AI models as personalized assistants, on edge servers to support intelligent driving decision making and enhanced on-board experiences. However, such cross-reality interactions may cause serious location privacy risks, as adversaries can infer AV trajectories by correlating the location reported when AVs request LBS in reality with the location of the edge servers on which their corresponding AI agents are deployed in virtuality. To address this challenge, we design a cross-reality location privacy protection framework based on hybrid actions, including continuous location perturbation in reality and discrete privacy-aware AI agent migration in virtuality. In this framework, a new privacy metric, termed cross-reality location entropy, is proposed to effectively quantify the privacy levels of AVs. Based on this metric, we formulate an optimization problem to optimize the hybrid action, focusing on achieving a balance between location protection, service latency reduction, and quality of service maintenance. To solve the complex mixed-integer problem, we develop a novel LLM-enhanced Hybrid Diffusion Proximal Policy Optimization (LHDPPO) algorithm, which integrates LLM-driven informative reward design to enhance environment understanding with double Generative Diffusion Models-based policy exploration to handle high-dimensional action spaces, thereby enabling reliable determination of optimal hybrid actions. Extensive experiments on real-world datasets demonstrate that the proposed framework effectively mitigates cross-reality location privacy leakage for AVs while maintaining strong user immersion within 6G-enabled vehicular metaverse scenarios.

Figures (8)

• Figure 1: Workflow of the proposed cross-reality location privacy protection framework based on hybrid actions in 6G-enabled vehicular metaverses. The sequentially numbered arrows depict the main processes: (a) The target AV at privacy risks moves and executes hybrid actions, i.e., the continuous location perturbation and discrete privacy-aware AI agent migration (Steps ① to ③). (b) The AV requests LBS and offloads AI agent tasks to edge servers, while malicious adversaries exploit the illegally obtained information of the target AV to launch inference attacks (Steps ④ to ⑧). (c) The on-board user obtains immersive metaverse applications without compromising location privacy (Step ⑨).
• Figure 2: Architecture of the proposed LHDPPO algorithm for the cross-reality Location Privacy Protection (LPP) environment.
• Figure 3: Left: Visualization of the AV trajectories and edge server locations. Right: Parameter and hyperparameter settings in this paper.
• Figure 4: Convergence performance of the proposed LHDPPO algorithm and the baseline methods. LR: learning rate. DS: denoising step
• Figure 5: Performance comparison between the proposed LHDPPO algorithm and the baseline methods.

Theorems & Definitions (1)

• Definition 1