This work aims to automatically generate or adapt elements of a first-person shooter (FPS) game based on learned mappings between level structures, game parameters, and gameplay outcomes. A convolutional neural network combines inputs from level design (the top-down FPS map) and game design (the parameters of the competing players' classes); it is trained on a vast corpus of automated playtest logs to predict the results of the match (the winner's kill ratio and match duration). This learned mapping allows for a fast (although not always accurate) search-based generation of levels or character classes towards desired gameplay outcomes such as a balanced long match.
Daniel Karavolos, Antonios Liapis and Georgios N. Yannakakis: "Using a Surrogate Model of Gameplay for Automated Level Design," in Proceedings of the IEEE Conference on Computational Intelligence and Games, 2018. PDF BibTex
Daniel Karavolos, Antonios Liapis and Georgios N. Yannakakis: "Pairing Character Classes in a Deathmatch Shooter Game via a Deep-Learning Surrogate Model," in Proceedings of the FDG Workshop on Procedural Content Generation, 2018. PDF BibTex
Antonios Liapis, Daniel Karavolos, Konstantinos Makantasis, Konstantinos Sfikas and Georgios N. Yannakakis: "Fusing Level and Ruleset Features for Multimodal Learning of Gameplay Outcomes," in Proceedings of the IEEE Conference on Games, 2019. PDF BibTex
Daniel Karavolos, Antonios Liapis and Georgios N. Yannakakis: "Learning the Patterns of Balance in a Multi-Player Shooter Game," In Proceedings of the FDG workshop on Procedural Content Generation in Games, 2017. PDF BibTex