Expert Knowledge for Real-Time Strategy Games
People
- Ben Weber
- Michael Mateas
- Josh McCoy (former member)
Overview
Our goal is to develop challenging AI opponents for real-time strategy games based on expert knowledge. Real-time strategy games include several challenging AI research problems including decision making with imperfect information, adversarial planning and opponent modeling. Our initial approach to overcome these challenges utilized a reactive planning agent, which partitions the problem space into domains of competence seen in expert play. This approach requires a human with expert knowledge of the game to author behaviors to counter the possible strategies in the game.
More recent work has made use of replays from expert players in order to reduce the number of behaviors that need to be specified by a human expert. One approach has been the application of case-based reasoning to build order in a real-time strategy game, which alleviates the human from needing to specify build orders in the agent. Rather, build orders are learned from a set of replays. We applied retrieval strategies using domain knowledge to constrain how the agent selects cases.
Our current focus is a statistical approach to modeling expert play. We have collected a large set of expert game traces and are applying machine learning techniques to automatically acquire domain knowledge. We are applying data mining techniques to strategy recognition as well as predicting opponent actions.
Publications
- B. Weber and M. Mateas, “Case-Based Reasoning for Build Order in Real-Time Strategy Games”, Artificial Intelligence and Interactive Digital Entertainment (AIIDE 2009), October 2009. pdf
- B. Weber and M. Mateas, “A Data Mining Approach to Strategy Prediction”, IEEE Symposium on Computational Intelligence in Games (CIG 2009), September 2009. pdf
- B. Weber and M. Mateas, “Conceptual Neighborhoods for Retrieval in Case-Based Reasoning”, International Conference on Case-Based Reasoning (ICCBR 2009), July 2009. pdf
Media
Current Project (EISBot):
http://eis.ucsc.edu/EISBot
Here's a flyer describing the project:
rts-1sheet.pdf
Arff files from the data mining paper:
http://users.soe.ucsc.edu/~bweber/traces
Here are several videos from the case-based reasoning papers that show the agent in action:
http://users.soe.ucsc.edu/~bweber/wargusreplays/wargusreplays.html
Results from earlier work on small-scale combat in Wargus:
wargus_results