Manipulating articulated objects with interactive perception

摘要: Robust robotic manipulation and perception remains a difficult challenge, in particular unstructured environments. To address this we propose to couple perception. The robot observes its own deliberate interactions with the world. These reveal sensory information that would otherwise remain hidden facilitate interpretation of perceptual data. demonstrate effectiveness interactive present skill for articulated objects. We show how UMan, our mobile platform, obtains kinematic model an unknown object. then enables perform purposeful manipulation. Our algorithm is extremely robust, does not require prior knowledge object; it insensitive lighting, texture, color, specularities, background, computationally highly efficient.