Motivated Reinforcement Learning Agents

Abstract

Motivated reinforcement learning agents combine a model of motivation with a reinforcement learning algorithm. The output of the motivation function becomes the reward signal input for the reinforcement learning algorithm. In this chapter we begin by providing a general motivated reinforcement learning algorithm that assumes an understanding of the computational models of curiosity based on interest and competence presented in Chap. 5. We then develop this general model using three specific reinforcement learning approaches.