Consistency-Driven Resource Allocation in Mobile Augmented Reality Systems

Mobile Augmented Reality (MAR) has emerged as a promising technology for multi-user collaborative applications, yet faces a critical challenge. In particular, different processing capacity in computational tasks leads to delay skew among users, thereby the spatial consistency of virtual objects is difficult to maintain. To address the challenge, we propose a novel joint optimization of task offloading and resource allocation, where optimization object is to minimize the delay skew among users. The optimization problem is NP hard and formulated as a Markov Decision Process. We develop a novel reinforcement learning algorithm, which integrate twin delayed deep deterministic policy gradient with parameterized dueling double deep Q-learning networks (TD3-PD3QN). Finally, simulation results show that the proposed scheme can achieves superior consistency performance.