A novel method that learns to reason about the 3D geometric relationship between objects, enabling robots to perform complex manipulation tasks that require understanding of object-centric representations.
