Meta-learning enhanced human–robot collaboration in visual communication design instruction

Visual icon-based human–robot interaction faces challenges of low efficiency and asynchronous responses in educational settings. This study proposes the Adaptive Graph-based Visual Teaching and Robot Interaction (AGVTRI) model for optimizing robot teaching networks. Integrating a Vision Transformer, graph neural network, and meta-learning, the model achieves adaptive visual feature extraction and relationship reasoning. Validated on Icons-50 and SVG datasets, it reaches 0.869 classification accuracy and 0.961 top-K accuracy. Physical tests on four educational robots show reduced interaction latency and improved action accuracy. The AGVTRI model enables high-fidelity mapping between visual cues and robot actions, supporting efficient, scalable intelligent teaching systems.