Cross-Modal Trajectory Prediction and Scene Understanding in Autonomous Driving Videos via Hierarchical Motion Encoding. JAAIR 2026, 5 (1).