Evolutionary Boss Improvement in Doom

Automating part of enemy balancing may save precious resources for developers, allowing them to focus on more creative tasks, while also enabling developers to create enemies that cater to specific kinds of players. In this paper, we adapt the "Evolutionary Boss Improvement" (EBI) method to the DOOM game, modifying the Cyberdemon enemy. This increases the method's generality by testing its effectiveness in another game genre. By implementing the EBI pipeline with appropriate adaptations to the encoding, decoding, crossover and mutation algorithms, we were able to improve the fitness of our enemies, which was evaluated by having a set percentage of remaining health after testing with bots. We asked anonymous volunteers to play the game against either the original Cyberdemon or our modified version (evolved using EBI) and to fill a survey about their experience, allowing us to assess their opinions about the boss fight. The version each player fought was randomly selected when starting the game, making it a double-blind experiment. The results indicate we created an individual with a similar feel to the original, maintaining player satisfaction, whilst also requiring less manpower from the developer.