To allow robots to achieve human-like dexterity, they not only need to be able to grasp objects, but also to move them within their hand. Consider, for example, writing with a pen: if all you could do was grasp it with your hand, and not move it between your fingers, you would not be able to write very well. This project is about designing grasps and motions of the hand to create useful motions of the object in the hand. The long-term goal is the development of control algorithms for robot hands to allow them to perform in-hand manipulation that allow them to function as assistants in human environments, which require human-like in-hand manipulation.
The proposed work is a systematic framework for designing sliding in-hand manipulation. A sliding-compliant grasp is both programmable and mechanically passive (and therefore stable). In the "place" phase of pick-carry-place, sliding can be used to gain extra dexterity, controlling motions that cannot be reliably controlled by moving the grasped object with position control. This is particularly true for recent human-interactive robots controlled with soft actuators with low positioning accuracy, like Baxter. Sliding may be inevitable anyway, and therefore designing a grasp for the place operation can be as important as designing for the carry. Controlled sliding also provides a means for compliant assembly without adding specialized hardware, like the RCC device. Design of compliance for sliding includes the choice of the location and the normal forces of the grasping fingers, subject to equilibrium and force closure constraints, to shape the frictional limit surface to achieve a desired mapping from disturbing forces (inertial or contact forces) to corrective slipping motion. An objective of the project is to demonstrate in-hand sliding manipulation for both quasistatic assembly tasks and for dynamic regrasping.
