Using deep neural networks, robots interpret 2D images to reconstruct a 3D understanding of their bodies. These networks, composed of three distinct modules, analyze the robot's movements and deduce ...