The document discusses using emotion as a reward signal in reinforcement learning. It proposes that appraisals of situations can generate emotions, and that emotions can then drive reinforcement learning. The model was implemented in the Soar cognitive architecture. Results showed that using emotion allowed the agent to learn tasks faster compared to standard reinforcement learning, and that incorporating mood led to even faster learning with less variability in performance. In the future, the researchers aim to explore more complex multi-agent scenarios using this approach.