Strategy games are complex environments often used in AIresearch to evaluate new algorithms. Despite the commonalities of most strategy games, often research is focused on one game only, which may lead to bias or overfitting to a particular environment. In this paper, we motivate and present STRATEGA - a general strategy games framework for playing n-player turn-based and real-time strategy games. The platform currently implements turn-based games, which can be configured via YAML-files. It exposes an API with access to a forward model to facilitate research on statistical forward planning agents. The framework and agents can log information during games for analysing and debugging algorithms. We also present some sample rule-based agents, as well as search-based agents like Monte Carlo Tree Search and Rolling Horizon Evolution, and quantitatively analyse their performance to demonstrate the use of the framework. Results, although purely illustrative, show the known problems that traditional search-based agents have when dealing with high branching factors in these games.
Github: https://github.com/GAIGResearch/stratega
Cite this work
@inproceedings{dockhorn2020stratega, author= {Dockhorn, Alexander and Grueso, Jorge Hurtado and Jeurissen, Dominik and Liebana, Diego Perez}, title= {{STRATEGA: A General Strategy Games Framework}}, year= {2020}, booktitle= {{AIIDE-20 Workshop on Artificial Intelligence for Strategy Games}}, abstract= {Strategy games are complex environments often used in AIresearch to evaluate new algorithms. Despite the commonalities of most strategy games, often research is focused on one game only, which may lead to bias or overfitting to a particular environment. In this paper, we motivate and present STRATEGA - a general strategy games framework for playing n-player turn-based and real-time strategy games. The platform currently implements turn-based games, which can be configured via YAML-files. It exposes an API with access to a forward model to facilitate research on statistical forward planning agents. The framework and agents can log information during games for analysing and debugging algorithms. We also present some sample rule-based agents, as well as search-based agents like Monte Carlo Tree Search and Rolling Horizon Evolution, and quantitatively analyse their performance to demonstrate the use of the framework. Results, although purely illustrative, show the known problems that traditional search-based agents have when dealing with high branching factors in these games.},
}