CAREER: Quantifying Humanlike Enveloping Grasps

Carnegie Mellon University, Pittsburgh PA

Investigators

Abstract

Over the past decade, our ability to produce graphical images has improved to the extent that we can create imaginary scenes that are virtually indistinguishable from reality. Digital humans have been called the last frontier in this march to graphical realism, and the area of human animation has also seen dramatic developments. Increasing use of motion capture data and new techniques for manipulating that data allow us to reproduce human motion at an extremely high level of fidelity. Graphically generated characters in video games and films can seem uncannily real. Pending the development of easy-to-use tools for directing digital humans, we should soon see digital humans as plausible user interfaces, and animated characters will become much more prevalent in education, demonstration, and training applications. If digital humans are the last frontier in realistic computer graphics, the last frontier in realistic digital humans is generating believable hand motion. Human hands are beautiful and complex mechanisms, amazing in their utility and adaptability. It is argued that it is our hands that make us human, and that hand evolution was a primary factor in the development of intelligence. Hand use in autonomous digital human characters, however, is generally quite unconvincing. Hands may be placed in a single frozen pose, and interaction between characters and objects is avoided when possible. The main problem is that geometric models of the human hand have far too much flexibility. This flexibility makes working with hands difficult even for trained animators, and it poses a tremendous challenge for creating autonomous characters that must interact with their environment. I believe that the key to making further progress in hand motion for digital characters is much more detailed consideration of the anatomy of the human hand. In analysis of human grasps, for example, critically important considerations include the amount of contact between finger pads, palm, and object; the ability of muscles to produce or resist task force; and the stabilization roles of fingers and muscles, yet none of these issues have been explored in grasp synthesis research in either the robotics or computer graphics communities. In pursuit of the goal of believable hand use for digital characters, we propose an anatomy-based model of human grasping. In particular, we propose a tendon-based quality measure for humanlike enveloping grasps, and we plan to evaluate this quality measure (1) for ability to discriminate between grasps, (2) as a predictor of grasp forces, and (3) for use in modeling grasp acquisition. Because of the strong emphasis on human anatomy, this research has the potential for additional impact outside graphics and animation in areas including ergonomics (tool design), robotics (robot hand design), and anthropology (research in human hand evolution and tool use). The educational portion of this proposal focuses on teaching and mentoring of undergraduates. The research ideas, techniques, and results will be incorporated into a course at Brown that attracts both un-dergraduate and graduate students, and will provide them with an opportunity to learn and experiment in a problem domain that is a nice mix of computational geometry and numerical optimization, grounded in human anatomy and supported by data. A special effort will be made to include undergraduate women in the research program, for example, through the CRA Distributed Mentor Program. Research results will include a library of example grasps and applied forces, as well as a tool for adapting the examples to new hand and object geometries. Once this research is published, the data and tools will be made available to other researchers on the web, and should serve as a useful resource for creating digital characters for education, entertainment, and training applications.