A.Salichs, “Emotion-Based Learning of Intrinsically Motivated Autonomous Agents living in a

Abstract- This paper presents the performance of autonomous agents living in a role-playing game, where social interaction is allowed. These agents are controlled by an emotion-based architecture where the control architecture has a motivational model, which performs homeostatic control of the internal state of the agent. The agents learn the behaviour selection using reinforcement learning algorithms where happiness/sadness of the agent are used as positive/negative reinforcement signals. The agent will use the standard Q-learning algorithm as the RL algorithm when the agent is not interacting with the other player. In the case of social interaction the agent must use a multiagent RL algorithm. Fear is used to prevent the agent choosing dangerous actions or being in dangerous states where non-controlled exogenous events, produced by external objects or other agents, could danger him. The conducted experiments through the role-playing game analyse the performance of the agent varying his “daring factor”. Index Terms- Autonomous personal social robots, emotions