Physical Manipulation

Physical Manipulation projects explore ML methods and systems that use purposeful interaction with the physical world to learn how to perceive, reason and act in the real world. Physical manipulation is intrinsically connected to robotic manipulation, using it as both a vehicle for learning and as an application domain. However, while robotic manipulation is usually concerned with performing tasks, physical manipulation focuses on aspects of broad interest to AI that require active exploration and interaction, like acquiring an intuitive understanding of physics or learning to establish cause and effect. The potential applicability of AI systems that understand the state of a physical environment and can predict its evolution goes well beyond robotics; it includes domains like mixed reality, ambient intelligence, image and video understanding and many more.