Reinforcement Learning Based Wireless Augmented Reality on Mobile Edge Computing

Augmented reality (AR) technologies provide a novel experience for users to interact with the real-world environment by integrating virtual content into real circumstances. However, wireless AR applications have trouble to ensure visual security in complex and various scenarios. In this paper, we propose a reinforcement learning based wireless AR, which adaptively revises the location of the virtual content to avoid obscuring real-world context. The AR devices adaptively make display strategies according to the partially observable Markov decision procession (POMDP). We incorporate mobile edge computing (MEC) to release the computing pressure and assist wireless AR device to make decisions. Finally, we compare the proposed algorithm to ε-greedy algorithm, which is widely used in reinforcement learning problems. The experiment results show that the proposed algorithm learns faster and shows better performance of avoiding obstructions. Besides, we discuss the assistance of MEC to improve the performance of learning convergence. Experiment results show that the assistance of MEC brings the advantage of efficient learning.