Abstract <p>Pedestrian trajectory prediction is a critical task in autonomous systems and smart city applications, where predicting the future movement of pedestrians based on observed data is essential for decision-making. In this work, we propose ARP-STGCN, a novel approach that simultaneously addresses trajectory prediction and the imputation of missing observations. By leveraging a Gated Convolutional Imputation Module and an Attraction–Repulsion Potential (ARP) model, ARP-STGCN not only handles incomplete trajectories but also effectively captures social interactions among pedestrians. Our method utilizes a spatio-temporal graph convolutional network (STGCN) to model dynamic interactions between pedestrians, while incorporating scene features to account for environmental context. Additionally, we introduce a Weighted Position Loss function to emphasize accurate prediction in later time steps, ensuring that cumulative errors are minimized. Our approach outperforms existing methods in terms of both prediction accuracy and computational efficiency, achieving significantly lower Average Displacement Error and Final Displacement Error across multiple benchmarks, including the ETH/UCY dataset and SDD dataset. Moreover, ARP-STGCN consistently demonstrates faster inference times compared to state-of-the-art methods, making it suitable for real-time applications. Code is available at <a href="https://github.com/fantot/ARP-STGCN/">https://github.com/fantot/ARP-STGCN/</a>.</p> Graphical abstract <p></p>