Working independently and in coordination with humans and other devices, intelligent robots have the potential for profound positive impact on society. To achieve this promise, robots must be fast, capable, and safe as they interact in complex, highly uncertain environments. In this context, the fundamental tasks of grasping and dexterous manipulation are critical: when robots touch their surroundings, they typically operate slowly and cautiously to avoid any accidental contact or damage. This National Robotics Initiative (NRI) research project will develop multi-purpose algorithms for dynamic grasping and manipulation to achieve human-like speed and effectiveness for a broad range of applications. For in-home assistive robots, for example, safety and speed are both important when manipulating either objects or people. Prosthetic devices would also benefit from reliable and dynamic semi-autonomy: too often the control burden is placed on the user, who has limited sensing and actuation through the prosthetic interface. Additionally, both small and advanced manufacturing industries that require interaction with multiple tools and parts may benefit from this research. Outcomes of this research may also benefit robotic use in safety-critical applications, such as in disaster relief efforts.
A fundamental challenge in dexterous manipulation lies in the complexity of the contact between robot and object. Frictional contact introduces mathematical challenges to the governing equations of motion, particularly discontinuities which result in a combinatorial number of hybrid modes. This combinatorial complexity frustrates standard algorithmic methods, typically limiting manipulation schemes to a strict sequencing of contacts, defined a priori. This research has two central hypotheses. First, that formal, optimization-based numerical methods can discover and verify simple (non-combinatoric) control approaches that will be both dynamic and robust for unstructured manipulation tasks. This simplicity parallels human control of finger forces. Second, by explicitly considering the dynamics of manipulation, this research will lead to more robust and capable approaches than purely static or quasi-static methods. The algorithms will be implemented and tested on a physical, multi-link arm and gripper and in a simulated environment.
This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.
