In a collaboration between Universities of Bremen and Malta, the Dungeons & Replicants project trained player models from 213 players interacting with a daily dungeon instance in the MMORPG Aion. Each of the 213 replicants (AI agents) was trained after one player playing one of the ten classes of Aion, and learned to map the game state (information about player, opponent and preceding skill) to action (skill usage) probabilities. The models were used to find the right level of challenge in a replicant vs environment (PvE) combat scenario against different types of enemies, and the balance between classes in a replicant vs replicant (1v1 PvP) scenario.
Johannes Pfau, Antonios Liapis, Georg Volkmar, Georgios N. Yannakakis and Rainer Malaka: "Dungeons & Replicants: Automated Game Balancing via Deep Player Behavior Modeling," in Proceedings of the IEEE Conference on Games, 2020. PDF BibTex
Johannes Pfau, Antonios Liapis, Georgios N. Yannakakis and Rainer Malaka: "Dungeons & Replicants II: Automated Game Balancing Across Multiple Difficulty Dimensions via Deep Player Behavior Modeling," in IEEE Transactions on Games, 2022. (accepted) PDF BibTex