Analysis of Artificial Intelligence Applied in Video Games

摘要

Deep reinforcement learning is growing faster than ever before in the artificial intelligence world. As its applications and accomplishments proliferate, the challenges it faces will also increase in number. In this paper, the author focuses on deep reinforcement learning being applied to the field of games and video games. S some of the achievements in various types of games, from 2D perfect information environment (board games) to 3D imperfect information environment (first person perspective games) and the various challenges that DRL faces when being applied to the field will be analyzed. Challenges such as sample efficiency, exploration & exploitation trade-off, and delayed & sparse rewards will be discussed. In addition, some solutions are suggested. The solutions might also be applied to the various game examples in order to improve their performance.