Computational Intelligence and Games in Practice

Transcription

1 Computational Intelligence and Games in Practice ung-bae Cho 1 and Kyung-Joong Kim 2 1 Dept. of Computer cience, Yonsei University, outh Korea 2 Dept. of Computer Engineering, ejong University, outh Korea {sbcho@cs.yonsei.ac.kr, kimkj@sejong.ac.kr}

2 Outline ession 1 General Overview on Game AI Competitions The First tep (General Guidelines) with Game AI Competition Portal ession 2 Making Your Entry for imulated Car Racing Championships Making Your Entry for tarcraft Championships ession 3 Making Your Entry for Ms. Pac-Man Competitions Change Parameters and Rules for Ms. Pac-Man Controller ession 4 uper Mario and BotPrize Competitions (Julian Togelius and Philip Hingston) Concluding Remarks

3 ession 1

4 General Overview on Game AI Competitions

5 Components of Games NPC 1 Game Logs Interface Games Contents Players Artificial Intelligence and Computational Intelligence for Games NPC = Non-Player Character 4

6 What are Game AI Competitions? Winner Games The Best AI Program AI Programs 5

7 Competitions at WCCI 2012

8 Competitions at IEEE CIG

9 Competitions at GECCO

10 Games 9

11 10

12 Game tates Controls Game tates Controls Game tates Controls Compete Each Other Competition Game erver Computer Program 1 Computer Program 2 Computer Program N 11

13 Game tates Controls Game tates Controls Game tates Controls Compete Each Other with Human Players Human Player 1 Human Player 2 Human Player M Competition Game erver Computer Program 1 Computer Program 2 Computer Program N 12

14 Game tates Controls Game tates Controls Game tates Controls Playing Alone Competition Game Competition Game Competition Game Computer Program 1 Computer Program 2 Computer Program N 13

15 Game tates Controls Game tates Controls Game tates Controls Playing Alone with Human Observers Human Observer 1 Human Observer 2 Human Observer M Competition Game Competition Game Competition Game Computer Program 1 Computer Program 2 Computer Program N 14

16 Generating Contents for Human Players Contents Generator 1 Contents Generator 2 Contents Generator N Competition Game Competition Game Competition Game Human Player 1 Human Player 2 Human Player M 15

17 The First tep: General Guidelines

18 chedule of the Competition Off-line Competition Competition Website Open ubmission Deadline Conference Day ubmit Your Entry with Documents (PPT) Usually, Live Competition + Result Announcement

19 General Questions hould I attend the conference to keep my entry for the game AI competition? hould I submit a paper for the conference? hould I register for the conference? hould I pay for the entry submission? 18

20 Resource for Game AI Competitions Main websites Car racing, Ms. Pac-Man, uper Mario, Unreal Tournament Game AI competition Portal Compiled and maintained by CILAB at ejong University Research papers Introductory papers for each competitions (published in IEEE TCIAIG) Competition paper tracks (IEEE CIG) 19

21 imulated Car Racing 20

22 Ms. Pac-Man 21

23 uper Mario 22

24 Unreal Tournament 23

25 Game AI Competition Portal 24

26 Find the Portal through earch Engine 25

27 Game AI Competition Portal (2) Doku WiKi Language: English, Korean (Partially upported) Purpose It is not easy to start the first step for the game AI competition Although there are lots of information on the web, they re not friendly to novice users We have edited the portal to give basic instructions and information to prepare your first entry for the competition Contents The basic information to build your own entry for the competition tep-by-step instructions, software, source code, videos and so on 26

28 27

29 Games? There are several game AI competitions Each game has different characteristics Board games Video games First-person shooting games Real-time strategy games Usually, you don t need money to buy game W You can find a lot of fantastic videos for the successful results of the competitions from web 28

30 29

31 30

32 31

33 Calendar 32

34 Links 33

35 Journals 34

36 IEEE CIG Conference Proceedings 35

37 Competition Paper CIG 36

38 Acknowledgements 37

39 ession 2

40 Making Your Entry for imulated Car Racing Championships

41 40

42

43 Playing by Yourself 42

44 Introduction 43

45 Competition tages Warm-Up Race alone Collect useful information about the track Qualifying Race alone Eight controllers that bridge the longest distances Race Race together F1-like scoring 44

46 Client-erver Model 45

47 ensors 19 Track ensors (ensor configuration is possible) Max=200m 36 Opponent ensors (10 degree) Position ensor 46

48 Actuators 47

49 Which Track for the Competition? 48

50 49

51 Install TORC 50

52 Patch the TORC for the Competition Access to the track data structure is not allowed in the competition 51

53 ample Client 52

54 Win32 Console Project Client.cpp 53

55 Where is a ample Car Control Code? 54

56 Code tructure imple Driver Car tates In tuck No getgear tuck Handling getteer Car Control getaccel 55

57 Functions in the ample Code 56

58 An Example of the wdrive Function float accel_and_brake = getaccel(cs); int gear = getgear(cs); float steer = getteer(cs); float clutch; clutching(cs,clutch); CarControl cc(accel,brake,gear,steer,clutch); 57

59 Modification on the ample Codes float accel_and_brake = getaccel(cs); int gear = getgear(cs); float steer = -1; float clutch; clutching(cs,clutch); CarControl cc(accel,brake,gear,steer,clutch); 58

60 59

61 Results 60

62 ample Controller 61

63 Code tructure (Revisited) imple Driver Car tates In tuck No getgear tuck Handling getteer Car Control getaccel 62

64 Tutorial 2 (Full Acceleration) accel=1.0; brake=0.0; 63

65 Tutorial 3 (Constant peed) double target_speed=100; double actual_speed=cs.getpeedx(); accel=0; if(target_speed-actual_speed>0) accel=(target_speed-actual_speed)/20; if(target_speed-actual_speed>20) accel=1; brake=0; 64

66 Tutorial 3 65

67 peed peed peed peed peed Control

68 Tutorial 4 (No teering) steer=0; accel=1; brake=0; 67

69 teering Wheel Ideas? Length of trajectory vs. peed reduction vs. Probability of out-oftrack (stuck)

70 Professional Driver

71 Tutorial 5 (Keeping the Middle) double target_pos=0; double actual_pos=cs.gettrackpos(); double angle=cs.getangle(); double error = target_pos - actual_pos; angle = angle + error; steer = angle/steerlock; 70

72 Tutorial 5 71

73 Making Your Entry for tarcraft Championships

74 73

75 74

76 75

77 Race 76

78 tarcraft Basics Units Resource Buildings Fog-of-War

79 Tech Tree for Terran 78

80 couting 79

81 Combat

82 Challenges of tarcraft 81

83 ummary 82

84 Install tarcraft Please upgrade to

85 Visual tudio 2008 Express 84

86 BWAPI 85

87 Chaos Launcher 86

88 Copy Files 87

89 ExampleAIModule ExampleAIModule.dll 88

90 Compiling the Bot 89

91 Run the ample Code 90

92 ingle Player 91

93 Automatic Loading AI Module BWAPI: Loaded the AI Module: bwapidata\ai\exampleaimodule.dll 92

94 Find the Closest Minerals and Mine It if ((*i)->gettype().isworker()) { } { } Unit* closestmineral=null; for(std::set<unit*>::iterator m=broodwar->getminerals().begin(); m!=broodwar->getminerals().end(); m++) if (closestmineral==null (*i)->getdistance(*m)<(*i)->getdistance(closestmineral)) closestmineral=*m; if (closestmineral!=null) (*i)->rightclick(closestmineral); 93

95 94

96 Find the Farthest Minerals and Mine It if ((*i)->gettype().isworker()) { { Unit* closestmineral=null; for(std::set<unit*>::iterator m=broodwar->getminerals().begin(); m!=broodwar->getminerals().end(); m++) if (closestmineral==null (*i)->getdistance(*m)>(*i)->getdistance(closestmineral)) closestmineral=*m; } if (closestmineral!=null) (*i)->rightclick(closestmineral); } 95

97 96

98 97

99 AIIDE tarcraft Competition 98

100 IEEE CIG

101 100

102 101

103 ession 3

104 Making Your Entry for Ms. Pac-Man Competitions

105 104

106 105

107 Maze Pill, Power Pill, Fruits Ghosts Randomness Characteristics tages 106

108 107

109 creen Capture 108

110 core Progress Best High core Best Avg core CEC 2007 WCCI 2008 CEC 2009 CIG 2009 CIG 2010 Event 109

111 CIG 2011 Results

112 Build Your Own Entry You have three options to start the Ms. Pac-Man competition. Option 1) JAVA code + WebPacMan.com - For the original version, you need to configure the color of objects and the position of the game screen. Option 2) C# code + Microsoft Revenge of Arcade - It supports automatic detection of game screen. However, you need to get the Microsoft Revenge of Aracade (from store). We bought the W from Amazon.com Option 3) You can simply download ICE Pambush 3 source code (JAVA + WebPacMan)and start from it. It is composed of about ten rules with A* search algorithm. It supports automatic detection of game screen. 111

113 Option 2 C# Version 112

114 Install Java DK 113

115 Install Java IDE 114

116 Download oftware Kit 115

117 1

118 Open the mspacman.ipr Project 117

119 Adjust creen Capture Position MsPacInterface.java 118

120 The Meaning of Left and Top (0,0) (left, top) 119

121 Change the Color-related Parameters 120

122 Final Results 121

123 ample Controller Agent.java 122

124 123

125 Changing Parameters and Rules for Ms. Pac-Man Controller

126 ICE Pambush 3 125

127 Preparation Please install JDK from ml Download ICE Pambush 3 source code from Unzip the source codes Compile the source codes (command prompt) javac main/mspacinterface.java encoding shift-jis Run Ms. PacMan through Execute the controller java main/mspacinterface 126

128 127

129 Tutorial 1 Is there any power pill? Find the trajectory to the nearest power pill using A* (avoiding ghosts) if(exist(gp.num_pp)) direction = as.run(gp.num_pp); // move to the closest power 128

130 Tutorial 2 if (Exist(gp.num_pill)) direction = as.run(gp.num_pill); // move to the closest pill 129

131 Tutorial 3 if (Exist(gp.num_pp)) direction = as.run(gp.num_pp); // move to the closest power pill else if (Exist(gp.num_pill)) direction = as.run(gp.num_pill); // move to the closest pill 130

132 Tutorial 4 if (Exist(gp.num_edible)) direction = as.run(gp.num_edible); // move to the edible ghost else if (Exist(gp.num_pp)) direction = as.run(gp.num_pp); // move to the closest power pill else if (Exist(gp.num_pill)) direction = as.run(gp.num_pill); // move to the closest pill 131

133 Tutorial 4 132

134 ession 4

135 uper Mario and BotPrize Competitions (Julian Togelius and Philip Hingston)

136 Concluding Remarks

137 Thank You