24
1 Realtime Strategy Task Force IEEE CIG 2015 StarCraft AI Competition Results Sehar Shahzad Farooq, In-Seok Oh, Man-Je Kim and Kyung-Joong Kim Cognition and Intelligence Lab (CILab) (http://cilab.sejong.ac.kr ) Dept. of Computer Engineering, Sejong University, Seoul, Republic of Korea [email protected], [email protected], [email protected], and [email protected]

Cig 2015 star craft ai competition results

Embed Size (px)

Citation preview

Page 1: Cig 2015 star craft ai competition results

1

Realtime Strategy Task Force

IEEE CIG 2015 StarCraft AI Competition Results

Sehar Shahzad Farooq, In-Seok Oh, Man-Je Kim and Kyung-Joong Kim

Cognition and Intelligence Lab (CILab) (http://cilab.sejong.ac.kr )

Dept. of Computer Engineering, Sejong University, Seoul, Republic of Korea

[email protected], [email protected], [email protected],

and [email protected]

Presenter
Presentation Notes
Hello Everybody, My name is Sehar farooq and I am one of the Co-organizers of IEEE CIG StarCraft AI Competition 2015 and Today I am going to present the results for this year’s competition. I am studying in Sejong University, south korea. We are also running the Cognition and intelligence laboratory there in Sejong university. You may find further details about our group and our research at the website. cilab.sejong.ac.kr
Page 2: Cig 2015 star craft ai competition results

2

Realtime Strategy Task Force

Organizers

Kyung-Joong Kim (Vice chair of RTS TF)

Sehar Shahzad Farooq (Assistant CILAB (Ph.D.))

In-Seok Oh (Former Professional Gamer in StarCraft 2)

Man-Jae Kim (Undergraduate Student)

Presenter
Presentation Notes
My Co-organizers are Prof. Kyung Joong Kim who has been the Vice Chair of the Real time Strategy game specially in Task force. In-Seok Oh, who has been a professional gamer in Starcraft. Down here is myself Sehar, I am a Ph.D Student and working as an assistant in CILab. While Mr. Man Jae kim is an undergraduate student in Sejong University, south Korea
Page 3: Cig 2015 star craft ai competition results

3

Realtime Strategy Task Force

Competition Background

Presenter
Presentation Notes
Before showing the results, lets talk a little about the Background of the Starcraft game
Page 4: Cig 2015 star craft ai competition results

4

Realtime Strategy Task Force

StarCraft Resource Management

Micro Management

Build Orders (Strategy)

Uncertainty (Fog-of-War)

Real-Time Response

Presenter
Presentation Notes
StarCraft is a RTS based game in which there are 5 important tasks. Frist is to collect the resources including mineral and gas and make the buildings to train the units. After we have the enough trained units, we need to attack the enemy. It depends upon the strategy (either attacking strategy or defending strategy. It also includes either to expand the base using the resources or training the units or attacking first . While attacking the enemies, we should scout the enemy positions using our trained units. All these things should be considered simultaneously. While uncertainty or fog of war is also important because we have to keep an eye on the enemy units because they may attack suddenly to our base. This information is helpful for making the decision about our strategy. It also depends on the unit’s power and range. While the main factor for the Real time response is the situation when we face the enemy’s units. That depends on the unit’s strength… either to attack the weak units or defend if the strong unit.
Page 5: Cig 2015 star craft ai competition results

5

Realtime Strategy Task Force

IEEE CIG StarCraft AI Competitions

Organizers # of Entries Winner

2015

Sehar Shahzad Farooq In-Seok Oh Man-Je Kim

Kyung-Joong Kim

14 ?

2014 Ho-Chul Cho In-Seok Oh

Kyung-Joong Kim 13 IceBot (Terran)

2013 Mike Preuss

Antonio Mora Garcia Tobias Mahlmann

8 Skynet (Protoss)

2012 Tobias Mahlmann Mike Preuss 10 Skynet (Protoss)

2011 Tobias Mahlmann Mike Preuss 10 Skynet (Protoss)

Presenter
Presentation Notes
IEEE CIG Started the Stracraft AI Competitions in 2011 where there was 10 entries for the competition…. Skynet was the winner in the Protoss race… There are Three type of races in Starcraft, The Protoss, Terran and Zerg. The same competition is repeated in 2012 with almost same entries…. Whine in 2013 there was only eight entries…Last year our lab conducted the competition with 13 entries and the winner was ICEBot because Skynet didn’t participate in 2014 and this year too…. This year there are 14 successful entries and you will now soon, about the winner of the game…..
Page 6: Cig 2015 star craft ai competition results

6

Realtime Strategy Task Force

Observations in StarCraft AI competitions since 2011

> AI

Empire of Protoss Race in AI World

> >

Presenter
Presentation Notes
Since the start of the IEEE CIG StarCraft AI Competition, it has been observed that the expert Human player is always better than the AI Bot…. The AI Bot couldn’t win against expert human player. However, in AI bot Vs AI bot Competition, Untill 2013 Protoss is strong race. But in 2014, Terran was the winner. There has been no participant for Zerg in CIG Starcraft competition for last three years…
Page 7: Cig 2015 star craft ai competition results

7

Realtime Strategy Task Force

2014 Results !!! (13 participants)

Rank Bot Name Main Contributor Race Affiliation Win Rate (%)

1 ICEBot Kien Nguyen Quang Terran Ritsumeikan University 83.06 2 Ximp Thomas Vajda Protoss Independent 78.06 3 LetaBot Martin Rooijackers Terran Maastricht University 68.47 4 AIUR Florian Richoux Protoss University of Nantes 66.11

5 UAlbertaBot 2013 David Churchill Protoss University of Alberta 60

6 WOPR Soren Klett Terran University of Bielefeld 56.53 7 MassCraft Dennis Soemers Protoss Maastricht University 55.14 8 NOVA Alberto Uriarte Terran Drexel University 38.89 9 MooseBot Adam Montgomerie Protoss University of Bristol 38.33

10 TerranUAB Filip Bober Terran Poznan University of Technology 34.03 11 BTHAI Johan Hagelback Terran Linnaeus University 31.53 12 NUSBot GU Zhan Protoss Institute of Systems Science, NUS 21.53 13 CruzBot Daniel Montalvo Protoss UC Santa Cruz 18.33

Presenter
Presentation Notes
In last year competition, The winner was Terran race with the win rate of 83%. The Bot name was ICEBot from Japan while the runner up was Ximp with 78% win rate. There were 13 participants in 2014.
Page 8: Cig 2015 star craft ai competition results

8

Realtime Strategy Task Force

2015 Competition

Page 9: Cig 2015 star craft ai competition results

9

Realtime Strategy Task Force

Race Newcomer Botname Contributor Affiliation

Zerg New ZZZBot Chris Coxe Independent

Zerg New Tscmoo_Z Vegard Mella Independent

Zerg New OverKill Sijia Xu Independent

Zerg New Salsa Pablo Garcia Sanchez University of Granada, Spain

Terran New Tscmoo_T Vegard Mella Independent

Terran New OpprimoBot Johan Hagelback Linnaeus University, Sweden

Terran New Tyr Simon Prins Utrecht University, Netherlands

Terran New Tequila Pablo Garcia Sanchez Politecnico di Torino, Italy

Terran LetaBot Martin Rooijackers Maastricht University, Nederland

Terran WOPR2014 Sören Klett University of Bielefeld, Germany

Terran NOVA Alberto Uriarte Drexel University, USA

Terran ICEStarCraftBot2014 Nguyen Duc Tung et al. Ritsumeikan University, Japan

Protoss NUSBot2014 Gu Zhan et al. National University of Singapore

Protoss Ximp Tomas Vajda Independent

Protoss AIUR2014 Florian Richoux Université de Nantes, France

Protoss UAlbertaBot2013 David Churchill University of Alberta, Canada

Entries (16 Participants, 14 Entries, 8 Newcomers, 2 crashed)

Presenter
Presentation Notes
This year, there were 16 entries for this competition. Among them 8 bots were new comers. While 2 bots couldn’t participate because of the incompatibility of their entries with our system. This time, there were 3 entries for the Zerg, 7 entries for Terran and 4 for the Protoss.
Page 10: Cig 2015 star craft ai competition results

10

Realtime Strategy Task Force

Basic Rules

• One vs. One match

• Full round robin • 30 rounds with 5 maps

• In total, 2730 games (30 minute time limitation)

• We used 20 machines for three days

• We had opened the maps to promote generalization

• No agreement on the release of source code > Recommended

• BWAPI (C++ or JAVA) and File I/O is allowed

Presenter
Presentation Notes
The basic rules includes the One vs. once match and its full round robin tournament with 30 rounds. We have fixed 5 maps this year and these maps were already opened to the participants. In total there were two thousand seven hundred and thirty games played in this competition, while, Each game is having 30 minute time limitation. We used 20 machines for three days. We recommended the participants to open the source code for future references. However, there are some participants who wanted their source code to not to open. We use Brood War API in Java or C++ language and the bots are allowed to read/write the I/O files.
Page 11: Cig 2015 star craft ai competition results

11

Realtime Strategy Task Force

Result Announcement!!!!! (3rd Rank)

Presenter
Presentation Notes
Lets see the results…. At first we will be opening the 3rd rank bot… Waiting…………..
Page 12: Cig 2015 star craft ai competition results

12

Realtime Strategy Task Force

3rd Rank

• Overkill (Zerg, Newcomer)

• Sijia Xu, Independent

• A* and Influence map AI technique

• Find the best attacking/ retreating position

• Script based AI, mainly using FSM with scalable architecture

• Main Attacking units are aircraft units

• Win Rates : 62.05%

Presenter
Presentation Notes
The Third rank bot is OverKill…….. He is a new comer in This year competition and he participated in Zerg race. The main contributor is Sijia Xu and the affiliation is independent. He used A* and influence map to find the best attacking or retreating position. He mainly used the Aircraft units named by Mutalisk. And the win rate is 62 point zero five percent.
Page 13: Cig 2015 star craft ai competition results

13

Realtime Strategy Task Force Overkill

Presenter
Presentation Notes
Th zergling are passing by the bunker and attacking the workers because if they attack the bunker, there is a no benefit to destroy the bunker…. The over kill make the spire (in a very short time) to make the mutalisk aircraft units. And the defense buildings The icebot is making the defense buildings but its time consuming. However, the overkill attack the base of the icebot quickly and destroy the base…
Page 14: Cig 2015 star craft ai competition results

14

Realtime Strategy Task Force

Result Announcement!!!!! (2nd Rank)

Page 15: Cig 2015 star craft ai competition results

15

Realtime Strategy Task Force

2nd Rank

• Tscmoo_Z (Zerg, Newcomer)

• Vegard Mella, Independent

• Powerful attacking strategy

• Avoid to confront enemy in weak situation

• Using guerrilla approach, find the steady weak enemy

• Win Rates : 73.59%

Presenter
Presentation Notes
The second rank bot is T sc Moo _ Z …….. He is a new comer in This year competition and he participated in Zerg and terran race But he won from the zerg race. The main contributor is Vegard mella and the affiliation is independent. He used the power attacking strategy. He avoid to confront the enemy in its weak situation. Since zerg has the most fastest units, so this bot used Guerrilla approach to find the steady weak enemy to attack. The win rate is seventy three point fifty nine percent.
Page 16: Cig 2015 star craft ai competition results

16

Realtime Strategy Task Force Tscmoo_Z

Presenter
Presentation Notes
The Tscmoo_z bot first scout the opponent base and then the enemy units in the attacking path….. It create the workers and defense buildings… Some When the enemies units
Page 17: Cig 2015 star craft ai competition results

17

Realtime Strategy Task Force

2015 Winner is …

Presenter
Presentation Notes
The 2015 Winner of the IEEE CIG StarCraft AI Competition is ……………………
Page 18: Cig 2015 star craft ai competition results

18

Realtime Strategy Task Force

Presenter
Presentation Notes
The 2015 Winner of the IEEE CIG StarCraft AI Competition is ……………………
Page 19: Cig 2015 star craft ai competition results

19

Realtime Strategy Task Force

1st Rank

• ZZZBot (Zerg, Newcomer)

• Chris Coxe, Independent

• 4 drone strategy Very fast sudden attack (Avg. time 5:52)

• Less unit producer (scarce force)

• Win Rates : 81.03%

Presenter
Presentation Notes
The winner is ZZZ Bot…. He is a new comer in This year competition and he participated in Zerg race. The main contributor is Chris Coxe, and the affiliation is Linnaeus University, sweden. He used the 4 drone strategy. He didn’t produce the workers but only the battle units to attack the enemy…. Since he used very fast sudden attack strategy, the avg game play time is around 6 minutes. The Win rate is 81. 03 %.
Page 20: Cig 2015 star craft ai competition results

20

Realtime Strategy Task Force ZZZBot

Presenter
Presentation Notes
He is not making the workers… instead making the buildings The opponent attack from the bunker, so instead of ignoring the bunker, he started attacking the bunker. Which means the strategy is failed. However,
Page 21: Cig 2015 star craft ai competition results

21

Realtime Strategy Task Force

Final Ranking

Page 22: Cig 2015 star craft ai competition results

22

Realtime Strategy Task Force

Rank Name Race Win (%) Comment 1 ZZZBot Zerg 81.03 Newcomer 2 Tscmoo_Z Zerg 73.59 Newcomer 3 Overkill Zerg 62.05 Newcomer 4 LetaBot Terran 61.54 3rd in CIG 2014 5 Ximp Protoss 60.26 2nd in CIG 2014 6 Tyr Terran 54.1 Newcomer 7 ICEStarCraft Terran 53.85 1st in CIG 2014 8 AIUR Protoss 53.33 3rd in CIG 2013 9 Tscmoo_T Terran 53.08 Newcomer 10 UAlbertaBot2013 Protoss 48.46 2nd in CIG 2013 11 WOPR Terran 46.41 12 NUSBot Protoss 21.79 13 OpprimoBot Terran 20.51 Newcomer 14 NOVA Terran 10

Presenter
Presentation Notes
This is the final ranking for the IEEE CIG STarCraft AI Comnpetiton 2015. It can be seen that the Zerg race is at the top rank. This is may be because of the reason that there were no participants in the last three years for the zerg race so the old and new comers couldn’t handle the zerg race. However, the winners used the opponent’s week point to quick attack in early strategy and got the ground. The NOVA bot has the lowest win ratio of 10%. This is because of the crashes in the bot. Approx there are about 50% crashes in games.
Page 23: Cig 2015 star craft ai competition results

23

Realtime Strategy Task Force

Remarks

• Source code of entries, results, and replays at competition website

• http://cilab.sejong.ac.kr/sc_competition/

• To be continued in 2016

• Please consider to create your new entries !!!

Page 24: Cig 2015 star craft ai competition results

24

Realtime Strategy Task Force

Thank you

RTS Task Force (Chair: Mike Preuss) http://gameai.itu.dk/rtsg/ StarCraft AI Competition Homepage http://cilab.sejong.ac.kr/sc_competition/