Table of Contents
CIG 2012 StarCraft RTS AI Competition
Over the last years, the AIIDE and CIG StarCraft competitions have seen quite some progress in the development and evolution of new StarCraft bots. Still, these would not be fit for beating expert human players, but the gap is seemingly getting smaller. StarCraft (by Blizzard) is one of the most popular RTS games of all time, is known to be extremely well balanced, and thus is an ideal candidate to test different AI approaches.
This competition is part of the Computational Intelligence in Games conference 2012.
|Submission deadline:||August 15, 2012|
|Conference:||September 12 - September 15, 2012|
If you have any questions or comments, please let us know via mail email@example.com.
|Bot Name||Main Contributor||Institution||Bot Race||Web Link|
|BroodwarBotQ||Gabriel Synnaeve||INRIA, Collège de France||Protoss||http://emotion.inrialpes.fr/people/synnaeve|
|Xelnaga||Ho-Chul Cho||Sejong University||Protoss|
|BTHAI||Johan Hagelbäck||Blekinge Institute of Technology||Terran||http://code.google.com/p/bthai/|
|SCAIL||Jay Young||The University of Birmingham, UK||Protoss||http://www.cs.bham.ac.uk/~jxy734/|
|AIUR||Florian Richoux||University of Nantes||Protoss||http://code.google.com/p/aiurproject/|
|Adjutant||Nicholas Bowen||University of Central Florida||Terran||http://code.google.com/p/adjutantbot/|
|IceBot||Wang Zhe||Ritsumeikan University||Terran|
|UAlbertaBot||David Churchill||University of Alberta, Canada||Protoss||http://webdocs.cs.ualberta.ca/~cdavid/|
|Nova||Alberto Uriarte||Drexel University||Terran||http://nova.wolfwork.com|
The results presentation as held on the CIG 2012 provides a good overview and some insights. The final standing is:
|Rank||Bot Name||Main Contributor||Win Rate||Avg Frames||1h Timeouts||Crahses||Timeouts|
Each of the 6 maps has been played on 675 times, 15 times by each bot-bot combination. Each bot thus played 810 games. We used 6 client computers, but most of the games (around 50%) were actually played on just 2 of them. As we did not have a network drive available, all reading/writing to disk by the bots was only local (on the computer currently used by the bot). We apologize for this.
The plot on the right (click to increase size) documents how the bots fared against the specific opponents. Note that Xelnaga is the only bot that regularly beats Skynet, whether Nova is especially strong against the UAlbertaBot. Some bots know their enemies…
Please note: there are at least 2 games where 3 players entered, which is to be considered as technical problem. However, we presume this did not affect the results very much.
Here are the 2 replays shown in the presentation: Xelnaga surprising Skynet with a targetted DTS attack on its probes, and Nova successfully luring UAlbertaBot's probes into its own base.
All the numbers have been computed using the tournament software provided by the AIIDE 2012 StarCraft competition team of the University of Alberta, Canada, namely Michael Buro, David Churchill, Jason Lorenz. Many thanks again!
The competition will use StarCraft Brood War 1.16.1. You must possess a legal copy of the game.
Bots for StarCraft shall use the Broodwar API, which provides hooks into StarCraft and enables the development of custom AI for StarCraft. Interfaces in many languages are available to query the current state of the game and issue orders to units. As of now, we expect all bots to use BWAPI version 3.7.3 which as also the version planned for the AIIDE tournament. However, there is a slight chance of an emergency bug fix to the BWAPI (such as happened last year) so that we need to switch to a newer version.
The following material is a bit outdated, as it was prepared by Ben Weber for the AIIDE StarCraft competition in 2010 (and not everything has been updated since then). However, it still provides a valuable starting point if you want to start designing and implementing a bot from scratch. Please note, that the referenced BWAPI versions are not used any more for our competition (we go with BWAPI 3.7.4 which has been issued in early July, 2012 and fixes some tournament-related issues with 3.7.3):
Due to lack of participation last year there will be only one track this year (formerly called “Track A”). Detailed rules are here.
This year, we will use the tournament software of the AIIDE competition, provided by Michael Buro, David Churchill and Jason Lorenz (many thanks, guys!). We will therefore have only one bracket and will play as many repetitions of round-robin as possible. Note that time constraints are quite hard and a large tolerance (1 min) applies only for the first frame to allow loading data and setting up.
The new version of the tournament software enables to use a persistent data store so that learning over multiple games gets possible.
We use a collection of 6 maps, 3 of which have 3 spawn points, and the other 3 have 6 spawn points. All maps have size 128×128 as some of the bots experience problems with other map sizes.
3 Player Maps
6 Player Maps
Tobias Mahlmann, IT University Copenhagen
Mike Preuss, Technische Universität Dortmund
For any comments or inquiries, please write to firstname.lastname@example.org
StarCraft®: Brood War®
Used with permission granted to CIG. Blizzard has consented to run the competition via the Broodwar API interface, which uses a back door of the game and is not in any form related to Blizzard. Thanks!
©1998 Blizzard Entertainment, Inc. All rights reserved. StarCraft, Brood War and Blizzard Entertainment are trademarks or registered trademarks of Blizzard Entertainment, Inc. in the U.S. and/or other countries.