Can robots learn to be social? Can they do that in a structured way? This project uses the DragonBot platform and state-of-the-art artificial curiosity algorithms to explore the possibility of robots learning to behave socially, similar to children. The robot reacts to people and receives internal rewards whenever the social interaction succeeds. Initially, the robot learns which behavior best initiates social interaction and later learns which behavior maintains that interaction for the longest period. The goal is to build a brain-inspired hierarchical curiosity-driven social behavior architecture, in which the robot autonomously learns a growing repertoire of social skills.