Home Courses Research Publications CV UAlbertaBot

Current Research Interests

Real-Time Strategy Game Combat AI - SparCraft

Real-Time Strategy Game Combat is a difficult area to apply traditional search or learning algorithms due to its extremely large state and action spaces and harsh real-time computational constraints. We have created SparCraft, an open source StarCraft combat simulator which comes packaged with several search algorithms for RTS Combat, including Alpha-Beta, UCT, and a new algorithm called Portfolio Greedy Search.

Related Publications:

• Portfolio Greedy Search and Simulation for Large-Scale Combat in Starcraft [2013]
  D. Churchill and M. Buro
  CIG 2013, Niagara Falls, Canada, 8 pages (Best Paper Winner)
• Fast Heuristic Search for RTS Game Combat Scenarios [2012]
  D. Churchill, Abdallah Saffidine, and M. Buro
  AIIDE, Stanford, 2012, pp.~112-117
• Incorporating Search Algorithms into RTS Game Agents [2012]
  D. Churchill and M. Buro
  AIIDE Workshop on Artificial Intelligence in Adversarial Real-Time Games, Stanford, 2012, 6 pages

Real-Time Strategy Game Build Order Planning

To win an RTS game you need an army, and to build an army efficiently and quickly you need a good build order. Our research in build order optimizatio has resulted in the first build order planning system capable of finding build orders in real-time under harsh tournament settings. All of the build orders in UAlbertaBot are planned in real time with this system with no hard coded build orders whatsoever.

Related Publications:

• Build Order Optimization in StarCraft [2011]
  D. Churchill and M. Buro
  AIIDE, Stanford, 2011, 6 pages

StarCraft AI Competitions

Since 2011 I have organized and competed in the AIIDE StarCraft AI Competition. For information please see the competition website.