Spatial-Aware VLA Pretraining through Visual-Physical Alignment from Human Videos Paper โข 2512.13080 โข Published 14 days ago โข 15