东北大学学报(自然科学版) ›› 2021, Vol. 42 ›› Issue (4): 478-483.DOI: 10.12068/j.issn.1005-3026.2021.04.004

• 信息与控制 • 上一篇    下一篇

基于强化学习的三维游戏控制算法

孟琭, 沈凝, 祁殷俏, 张昊园   

  1. (东北大学 信息科学与工程学院, 辽宁 沈阳110819)
  • 修回日期:2020-05-04 接受日期:2020-05-04 发布日期:2021-04-15
  • 通讯作者: 孟琭
  • 作者简介:孟琭(1982-),男,辽宁沈阳人,东北大学副教授.
  • 基金资助:
    国家重点研发计划项目(2018YFB2003502); 国家自然科学基金资助项目(62073061); 中央高校基本科研业务费专项资金资助项目(N2004020).

Control Algorithm of Three-Dimensional Game Based on Reinforcement Learning

MENG Lu, SHEN Ning, QI Yin-qiao, ZHANG Hao-yuan   

  1. School of Information Science & Engineering, Northeastern University, Shenyang 110819, China.
  • Revised:2020-05-04 Accepted:2020-05-04 Published:2021-04-15
  • Contact: MENG Lu
  • About author:-
  • Supported by:
    -

摘要: 基于强化学习,设计了一个面向三维第一人称射击游戏(DOOM)的智能体,该智能体可在游戏环境下移动、射击敌人、收集物品等.本文算法结合深度学习的目标识别算法Faster RCNN与Deep Q-Networks(DQN)算法,可将DQN算法的搜索空间大大减小,从而极大提升本文算法的训练效率.在虚拟游戏平台(ViZDoom)的两个场景下(Defend_the_center和Health_gathering)进行实验,将本文算法与最新的三维射击游戏智能体算法进行比较,结果表明本文算法可以用更少的迭代次数实现更优的训练结果.

关键词: 强化学习;深度学习;目标识别;Faster RCNN;DQN

Abstract: Based on reinforcement learning, an agent for three-dimensional first person shooting game(DOOM)was designed. The agent can move, shoot enemies and collect objects in the game environment. The proposed algorithm combines the Faster RCNN algorithm of deep learning and the Deep Q-Networks(DQN)algorithm of reinforcement learning, which can greatly reduce the search space of DQN algorithm and improve the training efficiency of the proposed algorithm. The experiments were carried out in two scenes(Defend_the_center and Health_gathering)of the virtual game platform(ViZDoom), and the proposed algorithm was compared with the state-of-the-art three-dimensional shooting game agent algorithm. The results show that the proposed algorithm can achieve better training results with fewer iterations.

Key words: reinforcement learning; deep learning; object detection; Faster RCNN; Deep Q-Networks(DQN)

中图分类号: