An Emotional Virtual Character: A Deep Learning Approach with Reinforcement Learning

• Gilzamir Gomes UECE
• Creto Vidal UFC
• Joaquim Cavalcante-Neto UFC
• Yuri Nogueira UFC

Resumo

We have developed an emotional agent based on reinforcement learning. The agent is a virtual character who lives in a three-dimensional maze world. Thus, we found that an emotional engine can induce the behavior of an agent after training. The training algorithm was the Asynchronous Advantage Actor-Critic (A3C). Experiments showed that the emotional engine and reinforcement learning produced coherent behaviors. This is an important step towards believable autonomous virtual characters.

Palavras-chave: Believable Virtual Characters, Emotion models, Deep Reinforcement Learning