Graph-Augmented Motion Encoding for Long-Term Human–Object Interaction Prediction in Egocentric Videos. JDIAS 2026, 1 (1).