CCS355 Neural Network & Deep Learning Unit II Notes with Question bank .pdf
Cig2014 starcraft_competition
1. 1
Realtime Strategy Task Force
CIG 2014 StarCraft AI Competition
Ho-Chul Cho, In-Seok Oh, and Kyung-Joong Kim*
Cognition and Intelligence Lab (http://cilab.sejong.ac.kr)
Dept. of Computer Engineering, Sejong University, Seoul, Republic of Korea
{chc2212@naver.com, ohinsuk@naver.com, kimkj@sejong.ac.kr}
2. 2
Realtime Strategy Task Force
Organizers
Kyung-Joong Kim
(Vice chair of RTS TF)
Ho-Chul Cho
(StarCraft AI Participant since 2011)
In-Suk Oh
(Former Professional Gamer in StarCraft 2)
8. 8
Realtime Strategy Task Force
Race
Newcomer
Botname
Contributor
Affiliation
Protoss
New
CruzBot
Daniel Montalvo
UC Santa Cruz, USA
Protoss
New
MooseBot
Adam Montgomerie
University of Bristol, UK
Protoss
New
MaasCraft
Dennis Soemers et. al
Maastricht University, Nederland
Protoss
New
NUSBot
Gu Zhan et al.
National University of Singapore, Singapore
Protoss
New
Ximp
Tomas Vajda
Independent
Protoss
AIUR
Florian Richoux
Université de Nantes, France
Protoss
UAlbertaBot2013
David Churchill
University of Alberta, Canada
Terran
New
LetaBot
Martin Rooijackers et. al
Maastricht University, Nederland
Terran
New
WOPR
Sören Klett
University of Bielefeld, Germany
Terran
New
TerranUAB
Filip Bober et al.
Poznan University of Technology, Poland
Terran
BTHAI
Johan Hagelbäck
Linnaeus University, Sweden
Terran
NOVA
Alberto Uriarte
Drexel University, USA
Terran
ICEBot
Nguyen Duc Tung et al.
Ritsumeikan University, Japan
Entries (13 Entries, 8 Newcomers)
9. 9
Realtime Strategy Task Force
Basic Rules
•One vs. One match
•Full round robin
•60 rounds with 20 maps (balanced and diverse maps)
•In total, 4680 games (1 hour time limitation)
•We used 20 machines for a week
•We do not open maps and entries, to promote generalization
•Agreement on the release of source code
•BWAPI (C++ or JAVA) and File I/O is allowed
11. 11
Realtime Strategy Task Force
3rd Rank
•LetaBot (Terran, Newcomer)
•Martin Rooijackers and M. Winands, University Maastricht, Nederland
•Terran Bot with many different strategies, some of which involve using a wall-in to stop early attacks. Uses depth first search and flood fill to calculate possible wall-in location(s).
•Win Rates : 68.47%
17. 17
Realtime Strategy Task Force
1st Rank
•ICEBot (Terran)
•Nguyen Duc Tung, Nguyen Quang Kien, Kawase Kiyohito, Yamamoto Takahiro, Lee Hyunchong, Awagakubo Ren, Ruck Thawonmas, Ritsumeikan University, Japan
•Potential flows, finite state machine, enemy strategy prediction which triggers adaptive strategy rules and a lot of other heuristic things are applied in this bot.
•Win Rates : 83.06%
21. 21
Realtime Strategy Task Force
Discussion
•A lot of newcomers
Some bots are created based on open-source UAlbertaBot by David Churchill and other entries
It helps to build CruzBot, MooseBot, NUSBot and TerranUAB
•Terran race are successful
•Improved micro-management skills
•Bots start to understand “terrain”
22. 22
Realtime Strategy Task Force
Remarks
•Source code of entries, results, and replays at competition website
•Please consider to create your new entries !!!
•To be continued in 2015
23. 23
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/