Human-Robot Interaction (HRI) is pivotal in to-day's technological landscape, as robots become increasingly integrated into various aspects of human activity, spanning industrial, service, and healthcare sectors. Effective collabo-ration between humans and robots is essential for optimizing productivity, safety, and user experience. HRI also raises ethical and social considerations, highlighting the need for safe and ethical robot deployment and fostering trust among users. Graph neural networks (GNNs) offer a cutting-edge approach in HRI, enabling robots to model complex relational data and capture nuanced social interactions. By leveraging GNNs, robots can recognize human activities, infer intentions, and adapt their behavior accordingly, fostering natural and engaging interactions. In this paper, we utilize Graph Convolution Networks (GCNs) for datasets like AIR-Act2Act, which provide rich information for teaching social skills to robots and serve as benchmarks for action recognition tasks. By leveraging the spatial and temporal relationships encoded in 3D skeletal data, GNNs empower robots to perceive and interpret human behavior with sophistication, facilitating seamless interactions in real-world settings.