Cross-Modal Knowledge Distillation for Long-Horizon Video Understanding With Hierarchical Motion Representations. JAAIR 2026, 1 (1).