The ability to reason for itself and make decisions for its own actions is a highly desirable ability for robots engaged in human-robot interaction (HRI). By being able to perceive the environment, select the optimal action w.r.t. the state of the environment and then observe and learn from the reactions that occur, a robot can be said to be fully autonomous in the interaction. There have been several different types of frameworks focusing on cognitive functionalities for autonomous robots, and one particularly promising venue for cognition focuses on the affect-driven cognitive frameworks. This paper presents an ongoing study focused on developing an affect-based architecture for the humanoid robot iCub with the purpose of having fully autonomous interactions with humans .