Seol mar théacs é seo: ShapeStacks: Learning vision-based physical intuition for generalised object stacking