Stratego Using Deep Reinforcement Learning and Search

University essay from KTH/Matematisk statistik

Author: Anton Falk; [2021]

Keywords: ;

Abstract: Algorithmic game theory is a research area concerned with developing algorithms for solving games using game-theoretic concepts, with many applications in areas where games are used as models to achieve knowledge. In the last decades, numerous game-playing bots have been created, and in many games, they outperform top humans. This project concerns theadaption of ReBeL, a bot developed in 2020 that plays imperfect-information games, to a variant of the board game Stratego. It is built on deep self-play reinforcement learning and search. Counterfactual regret minimization was used in this adaptation. The algorithm’s performance was evaluated by play against different versions of itself as well as two other, more primitive bots, with good results. This work shows that ReBeL is a generally applicable algorithm and that it has the potential to reach high performance in adaptations to other games too.

  AT THIS PAGE YOU CAN DOWNLOAD THE WHOLE ESSAY. (follow the link to the next page)