ScrewNet: Category-Independent Articulation Model Estimation From Depth Images Using Screw Theory