2013 fighting game artificial intelligence competition
DESCRIPTION
These are slides about the 2013 Fighting Game Artificial Intelligence Competition presented at The 2nd IEEE Global Conference on Consumer Electronics (GCCE 2013) on October 3, 2013.TRANSCRIPT
Fighting Game Artificial Intelligence Competition
Platform
Feiyu Lu, Kaito Yamamoto, Luis H. Nomura, Syunsuke Mizuno, YoungMin Lee,
and Ruck Thawonmas
Intelligent Computer Entertainment Laboratory,Ritsumeikan University
2013 IEEE 2nd Global Conference on Consumer Electronics (GCCE 2013), October 3, 2013
Video Games and Fighting Games Artificial Intelligence Game AI Competitions in 2013 Architecture and Game System Contest Rules and Result Conclusions and Future Work
Outline
2013 IEEE 2nd Global Conference on Consumer Electronics (GCCE 2013), October 3, 2013
Video Games and Fighting Games Artificial Intelligence Game AI Competitions in 2013 Architecture and Game System Contest Rules and Result Conclusions and Future Work
Outline
2013 IEEE 2nd Global Conference on Consumer Electronics (GCCE 2013), October 3, 2013
Electronic game
Human interaction with a user interface
Generate visual feedback on video device
Video Games and Fighting Games
2013 IEEE 2nd Global Conference on Consumer Electronics (GCCE 2013), October 3, 2013
Tekken Namco 41.4 million (as of 2011)
Street Fighter Capcom 34 million (as of 2012)
Mortal Kombat Midway/Warner Bros.
26 million (as of 2007)
Super Smash Bros. Nintendo 23.6 million (as of 2013)
Soul Calibur Namco 13.4 million (as of 2012)
Dead or Alive Tecmo 7.95 million (as of 2008)
Video Games and Fighting Games
2013 IEEE 2nd Global Conference on Consumer Electronics (GCCE 2013), October 3, 2013
Video Games and Fighting Games Artificial Intelligence Game AI Competitions now Architecture and Game System Contest Rules Conclusions and Future Work
Outline
2013 IEEE 2nd Global Conference on Consumer Electronics (GCCE 2013), October 3, 2013
Artificial + Computational Intelligence
Finite state machines / fuzzy state machines / neural networks
Represents NPC (non-player character)
Run competition to compare and find high performance AIs
Artificial Intelligence
2013 IEEE 2nd Global Conference on Consumer Electronics (GCCE 2013), October 3, 2013
Video Games and Fighting Games Artificial Intelligence Game AI Competitions in this year Architecture and Game System Contest Rules Conclusions and Future Work
Outline
2013 IEEE 2nd Global Conference on Consumer Electronics (GCCE 2013), October 3, 2013
Game AI Competitions
AIIDE StarCrafthttp://webdocs.cs.ualberta.ca/~cdavid/starcraftaicomp/
2013 IEEE 2nd Global Conference on Consumer Electronics (GCCE 2013), October 3, 2013
Game AI Competitions
CIG 2013http://ls11-www.cs.uni-dortmund.de/rts-competition/start
2013 IEEE 2nd Global Conference on Consumer Electronics (GCCE 2013), October 3, 2013
Game AI Competitions
Angry Birdshttp://aibirds.org/
2013 IEEE 2nd Global Conference on Consumer Electronics (GCCE 2013), October 3, 2013
Game AI Competitions
Physical Travelling Salesmanhttp://www.ptsp-game.net/
2013 IEEE 2nd Global Conference on Consumer Electronics (GCCE 2013), October 3, 2013
Game AI Competitions
Platformerhttp://platformersai.com/
2013 IEEE 2nd Global Conference on Consumer Electronics (GCCE 2013), October 3, 2013
Video Games and Fighting Games Artificial Intelligence Game AI Competitions now Architecture and Game System Contest Rules Conclusions and Future Work
Outline
2013 IEEE 2nd Global Conference on Consumer Electronics (GCCE 2013), October 3, 2013
Architecture and Game System (1/2)
Platform’s architecture
This illustration shows the interactions between AIs or humans
AI and human player can get information per frame (the smallest time unit of the game, 1 frame = 16.66 msec) from game, and send their commands to game to control their characters2013 IEEE 2nd Global Conference on Consumer Electronics (GCCE 2013), October 3, 2013
Architecture and Game System
2013 IEEE 2nd Global Conference on Consumer Electronics (GCCE 2013), October 3, 2013
Actions have to be chosen and input for each frame, or the character will not do anything
Information sent to AIs will have a simulated delay – to prevent AI players using an unfair reaction speed
Architecture and Game System (2/2)
2013 IEEE 2nd Global Conference on Consumer Electronics (GCCE 2013), October 3, 2013
Skill system
Players use some skills and moving actions to fight
In the game, a skill usually has 3 stages – Startup, Active and Recovery
Skill System (1/5)
2013 IEEE 2nd Global Conference on Consumer Electronics (GCCE 2013), October 3, 2013
Skill System (2/5)
2013 IEEE 2nd Global Conference on Consumer Electronics (GCCE 2013), October 3, 2013
Skill System (3/5)
Startup 1st stage
Prepare to attack No attack box so no
damage
2013 IEEE 2nd Global Conference on Consumer Electronics (GCCE 2013), October 3, 2013
Skill System (4/5)
Active 2nd stage
The small red box -- attack box -- can be seen on the character’s arm
In these frames, if the opponent’s attack hit box coincide with this attack box, the opponent will be damaged
2013 IEEE 2nd Global Conference on Consumer Electronics (GCCE 2013), October 3, 2013
Skill System (5/5)
Recover 3rd stage Recover to normal
status
Special frame: cancelable
Some skills can be used during cancelable frames
2013 IEEE 2nd Global Conference on Consumer Electronics (GCCE 2013), October 3, 2013
Video Games and Fighting Games Artificial Intelligence Game AI Competitions in 2013 Architecture and Game System Contest Rules and Result Conclusions and Future Work
Outline
2013 IEEE 2nd Global Conference on Consumer Electronics (GCCE 2013), October 3, 2013
Only one character “KFM-prototype” can be used. One game has 3 rounds. Each round has a 5-
second ready time and 60-second fighting time. The characters' positions will be reset when time
over, and the system starts a new round if it is not the last round.
The game score will be calculated as
And higher one is the game’s winner.
Contest Rules
2013 IEEE 2nd Global Conference on Consumer Electronics (GCCE 2013), October 3, 2013
BitterCoffeeFTGAI Nanjing ZhiFeng Multimedia Co., Ltd., China BloodMoon Nanjing XinYiWei Software Co., Ltd., China Buakaw Bangkok University, Thailand Death Rose FTGAI ? Hybrid Bangkok University, Thailand Kaiju Bangkok University, Thailand Kong Fu King Donghu college, Wuhan University, China SejongAI Sejong University, Republic of Korea T TAIST-Tokyo Tech, Thailand Tony Jaa Bangkok University, Thailand
Team List (Alphabetical order)
2013 IEEE 2nd Global Conference on Consumer Electronics (GCCE 2013), October 3, 2013
BitterCoffeeFTGAI
Fighting Game AI Competition
Qi Jin BachelorSoftware Engineer of Nanjing ZhiFeng Multimedia Co., Ltd.
E-mail : [email protected]
2013 IEEE 2nd Global Conference on Consumer Electronics (GCCE 2013), October 3, 2013
BitterCoffeeFTGAI
Intro of AI :
AI that uses a
“Random move action + Rule base attack skill”
Way to fight.
2013 IEEE 2nd Global Conference on Consumer Electronics (GCCE 2013), October 3, 2013
BitterCoffeeFTGAI
Way of thinking :Detective the distance to opponent , random select some skill
(prevent a rule-base defense) to attack according to the distance. Some as:
e.g:distance less than 100 and on the ground will cast THROW_A or THROW_B;
e.g:distance greater than 100 and on the ground will cast STAND_D_DF_FA,STAND_D_DF_FB or STAND_D_DF_FC(when energy > 300)
and so on.
2013 IEEE 2nd Global Conference on Consumer Electronics (GCCE 2013), October 3, 2013
BloodMoon AI
FTGAIC
2013.9.262013 IEEE 2nd Global Conference on Consumer Electronics (GCCE 2013), October 3, 2013
AI’s Outline
1、 A rule base AI
2、 Get information like distance, high and character’s status, and then select a fitness skill in case to attack.
Cheung Yi Chun Nanjing XinYiWei Software Co., Ltd.
E-mail : [email protected] IEEE 2nd Global Conference on Consumer Electronics (GCCE 2013), October 3, 2013
Buakaw: Thai Fight
Wisarat PhuranantanondAngkana SuwanjatupornKingkarn Sookhanaphibarn
School of Science and Technology
BANGKOK UNIVERSITY2013 IEEE 2nd Global Conference on Consumer Electronics (GCCE 2013), October 3, 2013
BuakawMain concept is based on game
experience heuristics.We analyzed our bot position
compared to location on screen and distance between players.
If our position is in the middle of screen that mean the distance between players is always in range for the fireball action.
2013 IEEE 2nd Global Conference on Consumer Electronics (GCCE 2013), October 3, 2013
BuakawWe divided the distance between
players into three categories.◦Far◦Mid◦Near
Within each category a rule-based action is used. For example:◦Far: use mainly a fire ball action◦Mid: check the opponent’s action to
determine the response.◦Near: use a counter attack strategy based on
opponent’s action. 2013 IEEE 2nd Global Conference on Consumer Electronics (GCCE 2013), October 3, 2013
Death Rose FTGAI
Lu liang / [email protected]
2013 IEEE 2nd Global Conference on Consumer Electronics (GCCE 2013), October 3, 2013
Rule base Defense AI
A good defender is better than a fury attacker !
So when there is no better tactic, try to defense more and wait counter chance.
According to the distance, guard the fireball and wait opponent move or escape / counter-attack.
2013 IEEE 2nd Global Conference on Consumer Electronics (GCCE 2013), October 3, 2013
Thanarat JaruenyingAngkana Suwanjatuporn
Worawat Choensawat
Hybrid
School of Science and TechnologyBANGKOK UNIVERSITY
2013 IEEE 2nd Global Conference on Consumer Electronics (GCCE 2013), October 3, 2013
HybridWe presents a rule-based strategy based
on fighting states.Two states are determined which are
Wining and Losing.Wining state: our HP > opponent’s HPLosing state: our HP < opponent’s HP
For each state, we can divided into two cases as follows:Far: distance of players over a threshold. Near: distance of players less then the
threshold. 2013 IEEE 2nd Global Conference on Consumer Electronics (GCCE 2013), October 3, 2013
HybridAfter we determined the state and the
case, we used rule-based method to define the action.
The rule based method is based on four parameters as follows:Enemy’s ActionEnemy’s status: Air, Stand, Crouch or downOur energy: to determine which attack action
to be used.Distance between players
We also add a cornering prevention to avoid our character get trap in the corner of screen.
2013 IEEE 2nd Global Conference on Consumer Electronics (GCCE 2013), October 3, 2013
KAIJUParanee Apiromsanee
Angkana Suwanjatuporn
Kingkarn Sookhanaphibarn
School of Science and Technology
BANGKOK UNIVERSITY
2013 IEEE 2nd Global Conference on Consumer Electronics (GCCE 2013), October 3, 2013
KAIJU We used an adaptive decision to determine a
character action based on the opponent’s attacks.
The adaptive decision can be divided into two states as follows: Counter attack Defense
The reason for dividing the decision into two states is below: There is a large amount of opponent actions. Based
on our experiments, counter attack actions can be used only for some actions . Therefore, we used a defense strategy for actions that counter attack actions cannot be used. 2013 IEEE 2nd Global Conference on
Consumer Electronics (GCCE 2013), October 3, 2013
KAIJU The decision is based on three main
parameters:Distance between playersEnemy skillOur HP
Simple is the best!
2013 IEEE 2nd Global Conference on Consumer Electronics (GCCE 2013), October 3, 2013
KONG FU KING
2013 IEEE 2nd Global Conference on Consumer Electronics (GCCE 2013), October 3,
2013
Intro
A simple AI that using rule base algorism.
When fighting in long range, copy enemy’s action.
When fighting in a close range, use a throw.
When fighting in a middle range, use fireball.
Lu Xiaoming
Donghu college Wuhan University
E-mail : xiswspxr@hotmail.
com
2013 IEEE 2nd Global Conference on Consumer Electronics (GCCE 2013), October 3,
2013
Fighting Game AI
Sejong University
2013 IEEE 2nd Global Conference on Consumer Electronics (GCCE 2013), October 3, 2013
Introduction
• AI Name : SejongAI
• Advisor – Kyung Joong Kim
• Developers – Dong Hyun Jang– Kyoung Jun Ahn
• Affiliation– Sejong University, South Korea
2013 IEEE 2nd Global Conference on Consumer Electronics (GCCE 2013), October 3, 2013
Outline
1. Using Skills
2. Strategies
2013 IEEE 2nd Global Conference on Consumer Electronics (GCCE 2013), October 3, 2013
Outline - Using Skills
• STAND_D_DF_FA• STAND_D_DF_FC• STAND_D_DB_BB• THROW_HIT• AIR_UB
2013 IEEE 2nd Global Conference on Consumer Electronics (GCCE 2013), October 3, 2013
Outline - Strategies
• AI’s first attack skill is “STAND_D_DB_BB”.
• And the major attack skill is “STAND_D_DB_BB”.
• When “STAND_D_DB_BB” hits opponent and knock down, we will use “AIR_UB” or “STAND_D_DF_FA” or “STAND_D_DF_FC”.
• Our Agent’s major strategy is knock down.2013 IEEE 2nd Global Conference
on Consumer Electronics (GCCE 2013), October 3, 2013
T
Fighting Game AI Competition
Mr.Phakhawat Sarakit (Toon)Master Degree Student
TAIST Tokyo-TechInformation and Communication Technology for Embedded Systems : ICTES
E-mail : [email protected]
2013 IEEE 2nd Global Conference on Consumer Electronics (GCCE 2013), October 3, 2013
AI’s OutlineMeasure-Sense-Action cycle .Measure distance X between opponent and my AI see enemy action and thing
about it then find the best way to fight back the enemy.Fininte-State Machine(FSM).
I don’t know anymore. I think it’s a simple and the best I had ever have
2013 IEEE 2nd Global Conference on Consumer Electronics (GCCE 2013), October 3, 2013
TONY JAAAyuvat TrongkronboonyavatAngkana SuwanjatupornWorawat Choensawat
School of Science and TechnologyBANGKOK UNIVERSITY
2013 IEEE 2nd Global Conference on Consumer Electronics (GCCE 2013), October 3, 2013
Methodology
We implemented a rule-based model by using a state of our character.
Character state can be divided into three states as follows: Fully Attack: the character losing HP at
minimal range. Attack & Defense: the character keeps losing
HP. Fully Defense: the character losing HP over a
maximum threshold.
2013 IEEE 2nd Global Conference on Consumer Electronics (GCCE 2013), October 3, 2013
Methodology
For each state, a rule-based method is used within the state.
For the rule-based method, four main parameters are considered. Distance between players Enemy’s action Energy to determine which skill to be used Enemy’s status
2013 IEEE 2nd Global Conference on Consumer Electronics (GCCE 2013), October 3, 2013
BitterCoffee
FTGAI BloodMooN BuaKaw DeathRose Hybrid KaijuKongFuKin
g SejongAI T TonyJaa
TotalScore
s Rank
BitterCoffee FTGAI
434
1347
560
1348
484
1302
443
1194
145
356
500
1384
310
759
120
358
337
811 8859 8
448 341 434 353 105 500 240 124 88
465 447 384 398 106 384 209 114 386
BloodMooN
565
1650
417
1247
386
1203
361
1093
0
21
720
2010
309
872
25
260
100
160 8516 9
551 413 500 343 0 705 295 121 52
534 417 317 389 21 585 268 114 8
BuaKaw
440
1650
582
1750
66
801
802
1878
688
1976
500
1533
238
1387
168
1186
561
17551391
6 5
658 586 735 566 576 500 499 634 739
552 582 0 510 712 533 650 384 455
DeathRose
515
1695
613
1795
933
2197
272
1103
561
1757
666
1957
243
777
21
324
261
10971270
2 7
565 500 264 400 505 666 282 90 535
615 682 1000 431 691 625 252 213 301
Hybrid
556
1803
638
1904
197
1119
727
1895
201
1219
694
2032
326
1080
186
734
360
11751296
1 6
646 656 433 600 336 591 267 231 355
601 610 489 568 682 747 487 317 460
Kaiju
854
2641
1000
2978
311
1021
438
1240
798
1778
1000
3000
484
1430
309
686
587
18581663
2 3
894 1000 423 494 663 1000 668 124 750
893 978 287 308 317 1000 278 253 521
KongFu King
500
1614
279
987
500
1466
333
1041
305
965
0
0
182
645
89
559
187
904 8181 10
499 294 500 333 408 0 122 212 401
615 414 466 375 252 0 341 258 316
SejongAI
689
2238
690
2125
761
1610
756
2220
673
1917
515
1567
817
2352
670
1473
190
11681667
0 2
759 704 500 717 732 331 877 268 534
790 731 349 747 512 721 658 535 444
T
879
2639
975
2738
835
1815
978
2673
813
2263
690
2311
910
2438
329
1524
666
19822038
3 1
875 878 365 909 768 875 787 731 653
885 885 615 786 682 746 741 464 663
TonyJaa
662
2186
899
2837
438
1242
738
1900
639
1822
412
1140
812
2393
809
1829
333
1015
1636
4 4
911 947 260 464 644 250 598 465 346 613 991 544 698 539 478 983 555 336
Final Ranking
Team name Total scores Ranking
T 20383 1
SejongAI 16670 2
Kaiju 16632 3
TonyJaa 16364 4
BuaKaw 13916 5
Hybrid 12961 6
DeathRose 12702 7
BitterCoffeeFTGAI 8859 8
BloodMooN 8516 9
KongFuKing 8181 102013 IEEE 2nd Global Conference on Consumer Electronics (GCCE 2013), October 3, 2013
1st Round http://youtu.be/wLWxg6IOq7g
2nd Round http://youtu.be/QgKwqfERo4w
3rd Round http://youtu.be/gamOUgUTpA8
2013 IEEE 2nd Global Conference on Consumer Electronics (GCCE 2013), October 3, 2013
Video Clips of the Game Betweenthe Runner-Up SejongAI (P1) and
the Winner T (P2)
Video Games and Fighting Games Artificial Intelligence Game AI Competitions in 2013 Architecture and Game System Contest Rules Conclusions and Future Work
Outline
2013 IEEE 2nd Global Conference on Consumer Electronics (GCCE 2013), October 3, 2013
We have implemented our fighting game AI platform and reported the first competition.
In the future, with the support from Dimps, we plan to design more sample characters and run a sequence of competitions at related international conferences.
You are welcome to participate in these competitions. http://www.ice.ci.ritsumei.ac.jp/~ftgaic/
Conclusions and Future Work
2013 IEEE 2nd Global Conference on Consumer Electronics (GCCE 2013), October 3, 2013