부산대학교 도서관

The Augmented Reality (AR) technology has been around since 1968. It has been used in various applications serving different domains. Especially in the gaming domain, AR made a huge success. Nevertheless, these AR games rarely used Machine Learning (ML) techniques. Instead, they used simple kind of Artificial Intelligence (AI) algorithms. Hence, these games are un-realistic and predictable and so players often got bored too quickly.The use of ML enhances the players’ experience dramatically and offers new and creative ways to playing games in AR. The main contribution of this paper is to propose a new approach that adopts Deep Reinforcement Learning (DRL) in AR games. In the proposed approach, a new algorithm is introduced to detect the player’s behavior and update its policy. To evaluate the proposed approach, the paper introduces a method that gathers synthetic data in a custom-made environment. To study the impact of the proposed approach, we compare DRL policies of the agents before and after applying the proposed algorithm. The results reveal that the accumulative rewards of the proposed algorithm performs better than the untrained policies.