Go-Blend is an implementation of the Go-Explore reinforcement learning algorithm which is able to generate agents that exhibit various behavioral and affective patterns as specified by its rewards functions. The system builds and maintains an archive of cells in an iterative fashion, with each cell representing a game state as well as the trajectory of actions, previous states, and affect demonstrations (such as traces) required to reproduce this state. Each cell is assigned a behavior reward (e.g. maximize in-game score), and an affective reward (e.g. model annotated arousal).
Matthew Barthet, Ahmed Khalifa, Antonios Liapis and Georgios N. Yannakakis: "Generative Personas That Behave and Experience Like Humans," in Proceedings of the Foundations on Digital Games Conference, 2022. PDF BibTex
Matthew Barthet, Ahmed Khalifa, Antonios Liapis and Georgios N. Yannakakis: "Play with Emotion: Affect-Driven Reinforcement Learning," in Proceedings of the International Conference on Affective Computing and Intelligent Interaction, 2022. PDF BibTex