In the Metaverse, real-time, concurrent services such as virtual classrooms and immersive gaming require local graphic rendering to maintain low latency. However, the limited processing power and battery capacity of user devices make it challenging to balance Quality of Experience (QoE) and terminal energy consumption. In this paper, we investigate a multi-objective optimization problem (MOP) regarding power control and rendering capacity allocation by formulating it as a multi-objective optimization problem. This problem aims to minimize energy consumption while maximizing Meta-Immersion (MI), a metric that integrates objective network performance with subjective user perception. To solve this problem, we propose a Multi-Objective Multi-Agent Evolutionary Reinforcement Learning with User-Object-Attention (M2ERL-UOA) algorithm. The algorithm employs a prediction-driven evolutionary learning mechanism for multi-agents, coupled with optimized rendering capacity decisions for virtual objects. The algorithm can yield a superior Pareto front that attains the Nash equilibrium. Simulation results demonstrate that the proposed algorithm can generate Pareto fronts, effectively adapts to dynamic user preferences, and significantly reduces decision-making time compared to several benchmarks.

Virtual reality (VR) technology has recently achieved notable success and been widely expected to interplay with more mobile multimedia services. To further enhance real-time immersive experience for VR applications, exploiting cooperative offloading among capable terminal devices should be emerged as an effective means. However, faced with diverse and surging mobile VR user requests, terminal-assisted offloading needs to support comprehensive cached content, ultra-low latency delivery, and continuous energy provisioning, to guarantee stringent quality of service requirements, which poses a critical challenge for resource-constrained terminals. Hence, this paper proposes a Cooperative Offloading framework for Terminal Clusters (named CO-TC), in which VR terminal clusters form several cooperation groups for sharing cached field of view (FoV) tiles and available computing resources to cooperatively perform FoV rendering and content delivery. To maximize energy efficiency in CO-TC, an optimization problem is formulated to jointly decide the task offloading and computing resource utilization. An intelligent offloading scheme is designed based on multi-agent reinforcement learning (MARL) specially using agent relation feature graph embeddings. Moreover, we theoretically prove the permutation invariance and convergence of the proposed algorithm and derive the optimal observation range of the agent to balance the performance gain and interaction overhead in the distributed MARL frame. Finally, simulation results show that the proposed offloading scheme outperforms other baselines in terms of VR service performance, including latency, energy consumption, and energy efficiency.

As a critical and prevalent service in future mobile networks, virtual reality (VR) is latency-sensitive and power-hungry, bringing out the optimization problem of trade-off among power saving, delay, and resource utilization. Content caching and render offloading are deemed as promising solutions to meet the stringent requirements of VR on data transmission speed and end-to-end latency. In this article, we propose a novel distributed computing framework based on multi-agent deep deterministic policy gradient (MADDPG) for joint optimizing terminal-cooperative caching and offloading for VR tasks. Since the individual VR user can hardly reach the optimal actions based on its limited local observed states and samples, MADDPG with centralized training and distributed execution is exploited to solve the above challenge. In addition, the generative adversarial network (GAN) is introduced to obtain experience-enhanced agents in the offline training phase and to achieve an optimal allocation to minimize energy consumption in the online inferring phase. The Nash equilibrium is proven in the case that the distribution of finite real VR data samples is well imitated and complemented by GAN. Numerical results demonstrate that our algorithm has significant superiorities in terms of convergence performance and energy consumption over other benchmarks.