Using Go-Explore for efficient automatic playtesting of modern tabletop games

Automated Game Design Learning (AGDL) is an emerging field in AI research with the purpose of learning game design models through playing. The current strategy is to play out the full game in thousands of iterations, which can be impractical for complex games with large state space and computationally expensive forward models. More efficient exploration of the state space can help improve that. Go-Explore is a recent exploration paradigm that outperforms previous state-of-the-art by a large margin. The proposed research will focus on applying Go-Explore to improve the efficiency of automated playtesting of tabletop games by using an archive of interesting game states to reduce the time needed for self-play. The research will be primarily conducted within the TAG framework and aim to be game-agnostic. On successful completion, this research will improve game development cycles, resulting in higher-quality games, and potentially give unique insights into the game design process.