We study geometric augmentations for robustness in ABC tasks, demonstrating improvements across multiple noisy benchmarks.