(3.210.184.142) 您好!臺灣時間:2021/05/16 02:20
字體大小: 字級放大   字級縮小   預設字形  
回查詢結果

詳目顯示:::

: 
twitterline
研究生:陳朝政
研究生(外文):Chao-Cheng Chen
論文名稱:以增強式學習法設計之智慧型瑪利歐控制器
論文名稱(外文):An Intelligent Mario Controller Based on Reinforcement Learning
指導教授:蔡志忠蔡志忠引用關係
指導教授(外文):Jyh-Jong Tsay
口試委員:蔡志忠蔡源斌洪紹鑫
口試委員(外文):Jyh-Jong TsayYuan-Bin TsaiShao-Shin Hung
口試日期:2011-07-27
學位類別:碩士
校院名稱:國立中正大學
系所名稱:資訊工程研究所
學門:工程學門
學類:電資工程學類
論文種類:學術論文
論文出版年:2011
畢業學年度:99
語文別:英文
論文頁數:31
中文關鍵詞:電玩遊戲超級瑪利歐兄弟智慧型控制器競賽增強式學習法
外文關鍵詞:gamesSuper Mario BrosAI controllerscompetitionsreinforcement learning
相關次數:
  • 被引用被引用:0
  • 點閱點閱:268
  • 評分評分:
  • 下載下載:14
  • 收藏至我的研究室書目清單書目收藏:0
  近年來人工智慧在電玩遊戲上的應用是個越來越引人注目的議題,Mario AI Competition自從2009年起由IEEE Games Innovation Conference以及IEEE Symposium on Computational Intelligence and Games來舉辦,此競賽將任天堂的經典平台遊戲超級瑪利歐兄弟修改成為benchmark軟體讓人們可以創造自己的智慧型控制器來操縱瑪利歐克服各種不同的關卡。在此篇論文中我們使用機器學習中的一種學習方法,增強式學習法來建構智慧型瑪利歐控制器並使它在一個複雜的遊戲環境中學習,接著我們訓練控制器使它有足夠的能力以因應各種不同的難度及類型的關卡。我們在控制器的發展階段中謹慎的設計了新的狀態以及動作的定義來縮減增強式學習法的search space,這種設計可以將增強式學習法轉變成一種更加符合online learning需求的演算法。
Artificial Intelligence for computer games is an interesting topic in recently years. In this context, Mario AI Competition modifies a Super Mario Bros game to be a benchmark software for people who program AI controller to direct Mario and make him overcome the different levels. This competition was handled in the IEEE Games Innovation Conference and the IEEE Symposium on Computational Intelligence and Games from 2009. In this paper, we use the Reinforcement Learning which is one of learning mechanisms in Machine Learning area to construct a Mario AI controller that learns from the complex game environment. Then we train the controller to grow stronger for dealing with several difficulties and types of levels. In controller developing phase, we design the states and actions cautiously to reduce the search space and make the Reinforcement Learning become much suitable for the requirement of online learning.
Contents
Chapter 1 Introduction 1
1.1 Motivation 1
1.2 Organization of the Thesis 2
Chapter 2 Related Works 3
Chapter 3 Benchmark 5
3.1 Control 5
3.2 Mario 5
3.3 Enemies 6
3.4 Environment 8
3.5 Available APIs and Important Constraints 11
Chapter 4 Reinforcement Learning and Q-Learning 14
4.1 Q-Learning 14
4.2 Q-Learning’s Disadvantage 17
Chapter 5 Mario Controller Design Using Q-Learning 18
5.1 State and Action Task 18
5.2 Rewards and Parameters 22
Chapter 6 Experiment 25
Chapter 7 Conclusions 29
7.1 Summary 29
7.2 Future Work 29
Bibliography 31

List of Figures
Figure 2.1 : Modular reinforcement learning architecture 4
Figure 3.1 : Different types of Mario mode 6
Figure 3.2 : Mario’s Enemies 8
Figure 3.3 : Bricks in level 10
Figure 3.4 : The process of Mario’s power-up 10
Figure 3.5 : Landforms and obstacles in Mario’s world 11
Figure 3.6 : Some APIs in Environment Java interface of the benchmark 12
Figure 3.7 : The visualization of sensors in ovservation API 12
Figure 4.1 : Standard Reinforcement Learning architecture 14
Figure 4.2 : Q-Learning Algorithm 16
Figure 5.1 : The feedback's transmittal of Attack Task in Q-LearnerV1 21
Figure 5.2 : Attack Task 23
Figure 5.3 : Pass Through Task 23
Figure 5.4 : Avoidance Task 24
Figure 5.5 : Upgrade Task 24
Figure 6.1 : Win rate of Q-LearnerV1 and Q-LearnerV2 27
Figure 6.2 : Learning curve of Q-LearnerV1 and Q-LearnerV2 27

List of Tables
Table 6.1 : Result of the CIG conference in 2009 26




[1] Super Mario Bros. [Online]. Available: http://en.wikipedia.org/wiki/Super_Mario_Bros.
[2] Platform game. [Online]. Available: http://en.wikipedia.org/wiki/Platform_game
[3] J. Togelius, S. Karakovskiy, and R. Baumgarten, The 2009 Mario AI Competition, in IEEE Congress on Evolutionary Computation, 2010.
[4] R. S. Sutton and A. G. Barto, Reinforcement Learning: An Introduction, The MIT Press, 1998.
[5] J. Togelius, S. Karakovskiy, J. Koutn´ık, and J. Schmidhuber, Super Mario Evolution, in CIG’09: Proceedings of the 5th international conference on Computational Intelligence and Games. Piscataway, NJ, USA: IEEE Press, 2009, pp. 156–161.
[6] A* search algorithm. [Online]. Available: http://en.wikipedia.org/wiki/A*_search_algorithm
[7] I. Millington and J. Funge, Artificial Intelligence for Games. Morgan Kaufmann Pub, 2009.
[8] Bojarski and C. B. Congdon, REALM: A Rule-Based Evolutionary Computation Agent that Learns to Play Mario, Proceedings of the 6th international conference on Computational Intelligence and Games. Dublin, Ireland, IEEE Press, 2010, pp. 83–90
[9] G. A. Rummery and M. Niranjan, On-line Q-learning using connectionist systems, Engineering Department, Cambridge, University, 1994
[10] C.J. Hanna, R.J. Hickey, D.K. Charles and M.M. Black, Modular Reinforcement Learning Architectures for Artificially Intelligent Agents in Complex Game Environments, Proceedings of the 6th international conference on Computational Intelligence and Games. 2010, pp. 380–387
[11] Infinite Mario bros. [Online]. Available: http://www.mojang.com/notch/mario/index.html
[12] C. Watkins, Learning from Delayed Rewards, Cambridge University, England, 1989.

QRCODE
 
 
 
 
 
                                                                                                                                                                                                                                                                                                                                                                                                               
第一頁 上一頁 下一頁 最後一頁 top