Manipulator-Independent Representations for 

Visual Imitation