Robots have been gradually leaving laboratory and factory environments and moving into human populated environments. Various social robots have been developed with the ability to exhibit social behaviors and collaborate with non-expert users in different situations. In order to increase the degree of collaboration between humans and the robots in human–robot joint action systems, these robots need to achieve higher levels of interaction with humans. However, many social robots are operated under teleoperation modes or pre-programmed scenarios. Based on homeostatic drive theory, this paper presents the development of a novel collaborative behavior controller for social robots to jointly perform tasks with users in human–robot interaction (HRI) experiments. Manual work during the experiments is reduced, and the experimenters can focus more on the interaction. We propose a hybrid concept for the behavior decision-making process, which combines the hierarchical approach and parallel-rooted, ordered, slip-stack hierarchical architecture. Emotions are associated with behaviors by using the two-dimensional space model of valence and arousal. We validate the usage of the behavior controller by a joint attention HRI scenario in which the NAO robot and a therapist jointly interact with children.