We introduce a novel trajectory predictor that considers social interactions among agents, maintaining spatial-temporal information over an extended temporal horizon. It achieves high accuracy, generalisability, and outperforms recent SOTA algorithms on NGSIM and HIGHD Datasets