Don't Blind Your VLA: Aligning Visual Representations for OOD Generalization - Explained Simply | ArXiv Explained