Affective Reinforcement Learning: A Taxonomy and Survey
Published in Arxiv (Preprint). Submitted to IEEE Transactions on Affective Computing, 2025
Recommended citation: Barthet, M., Khalifa, A., Liapis, A., & Yannakakis, G. N. (2025). Affective Reinforcement Learning: A Taxonomy and Survey. Unpublished.
This paper surveys the current state-of-the-art of reinforcement learning methods and principles applied to affective computing across a wide variety of domains. We review this nascent field, which we refer to as affective reinforcement learning, that interweaves core RL components within the affective loop. Specifically, we survey the use of affective information across the three major components of the reinforcement learning (RL) paradigm: a) shaping a reward signal, b) driving the action policy, and c) forming part of the state representation. We introduce a taxonomy of different terms for affective RL, and we conclude by discussing the current limitations of the framework as well as several open and promising research directions within this emerging area.