The assertive actions done by social robots in education have to consider cognitive and emotional aspects of students. The document presents the HRS-EDU architecture to control the social robot behavior in learning environments. Concepts such as psychological theory of flow, scaffolding educational strategy and software agents were considered to design the BDI agent architecture (Beliefs, Desires, Intentions). The computational implementation was made in BDI-BESA which is an integrated development environment designed by the SIRP research group of the Pontificia Universidad Javeriana and integrated into the Baxter robot. To validate the architecture, three phases were implemented: remote control of the robot while the subjects solve the Jumper problem, the subjects try to solve the problem without the robot support and finally, the robot gives autonomous support during the problem-solving process. The validation phases were developed with 15 children ranged between 10–14 years old and who came from three schools placed in Bogotá. Through mixed methods research strategy, the events of Baxter, the reactions of the subjects, and the performance of the architecture were analyzed. The results suggest that the design of BDI goals for the control of the robot for the actions that give emotional and cognitive support to the subjects, encourages the learning of problem-solving strategies.