Abstract:
Tool manipulation is an essential human skill. It extends our manipulation capability beyond the capability of the biological hand, and is a defining feature of many important jobs centered on physical interaction with the real world. Yet, wielding a tool is drastically different from generally grasping an object. The prime examples are pens and scissors. When holding a pen to write and holding scissors to cut, our hand would reach very particular configurations, which significantly differ from the hand configurations for grasping these objects to transport them. In this work, we explore the following question: how to ensure that a robotic hand can successfully wield tools in the hand’s design stage? To tackle this question, we developed a computational design framework, which first takes advantage of the contact and kinematic constraints in tool-wielding to restrict the hand’s feasible design space, and then solves a highly constrained, multi-objective design optimization problem by approximating the feasible design space’s topology with a sampling-based strategy. In preliminary experiments, our framework generated thousands of hand designs. Nearly all of them succeeded in wielding tools, and their performance landscapes were mapped across varying design parameters. The preliminary results demonstrate our framework’s potential in efficiently designing tool-wielding hands, as well as providing valuable understandings of the feasible design space.

Committee:
Prof. Nancy Pollard
Prof. Jean Oh
Prof. Matthew Mason
Dominik Bauer