GAP: Geometric Anchor Pre-training for Data-Efficient Visuomotor Learning of Manipulation Tasks Paper • 2605.15836 • Published May 15 • 1