Recognizing human intentions to make robots able to collaborate with humans has been a high-interest research topic in recent years. In particular, non-verbal communication methods can provide useful approaches to reach a fluid and natural interaction between humans and robots. This article investigates the role of hand gesture recognition in human-robot interaction (HRI) scenarios, in which a human communicates with a robot using hand gestures. In particular, we emphasize the training and calibration of YOLOv2 (You Only Look Once) for real-time hand gesture recognition. We use YOLOv2 to detect five distinct hand gestures (index finger, full open hand, fist, thumb up, and peace sign). Then, we program 6 \(^\circ \) C of freedom (DoF) manipulator educational robot to perform a specific task based on the recognized gesture. We explore the creation of user-specific models, which achieve a mean average accuracy of up to 99.02%, and user-general models, with a mean average accuracy of up to 90.52% respectively. The outcomes of this research could significantly influence the HRI field and herald new developments enhancing the efficiency and fluidity of human-robot non-verbal communication.