Rogue-Like Games As A Playground For Artificial Intelligence - Evolutionary Approach

Rogue-likes are difficult computer RPG games set in a procedurally generated environment. Attempts have been made at playing these algorithmically, but few of them succeeded. In this paper, we present a platform for developing artificial intelligence (AI) and creating procedural content generators (PCGs) for a rogue-like game Desktop Dungeons. As an example, we employ evolutionary algorithms to recombine greedy strategies for the game. The resulting AI plays the game better than a hand-designed greedy strategy and similarly well to a mediocre player - winning the game 72% of the time. The platform may be used for additional research leading to improving rogue-like games and general PCGs.