RageVLA: Augmentation-Trained Perception and Action for State-of-the-Art Robustness on LIBEROwith Smarak Kanjilal1 versionW_mbtgccebv1 · current