Dexterous manipulation is a Grand Challenge in robotics today. Dexterity is required for robots working in the home, in space, for medical applications, disaster scenarios, and flexible manufacturing. Robots in these environments must be able to maneuver, lift, and handle objects, use them as tools, and competently transfer objects from one secure hold to another in a wide variety of uncertain situations. Such operations are very difficult for robots and present a barrier to their wider adoption. Recent advances in rapid prototyping and digital manufacturing have made it possible to design, manufacture, and test custom robot manipulators at a very rapid pace. One aim of this project is to explore and understand how these new technologies may be used to design a radically different type of robot manipulator that is customized for dexterous maneuvers. A second aim of the project is to increase our fundamental understanding of dexterity itself. The end goal is to enable robots to live up to their potential. In so doing, robots will begin to play an increasingly important role in our daily lives, and will inspire young students to prepare for, and pursue STEM careers.
The key research goal of this project is to formalize novel mathematical models and computational approaches to co-design mechanical structures and control policies for dexterous robotic manipulation tasks. Through this approach, control policies can be made significantly simpler, and mechanical features such as joint stops and compliance can passively improve the robustness of the manipulation tasks while reducing sensing and actuation requirements. Grasp nets, which capture specific families of manipulation capabilities observed in human performances, focus the efforts of this project and guide the design processes that are developed.
