CAREER: Primitives and Policies for Complex Behavior in Human and Robotic Hands

Arizona State University, Scottsdale AZ

Investigators

Abstract

Each human fingertip has approximately 2000 tactile sensors. Stimulation of these sensors triggers reactive grip responses that are mediated by the spine. In comparison to the dexterous capabilities of the human hand, robotic manipulation capabilities in unstructured environments are crude. When controlled by a human operator, robotic manipulators are further limited by restricted information flow (command and sensing) at the human-machine interface. All human-machine systems, from telesurgery robots to neuroprostheses, must address the critical issue of communication delays which can range, depending upon the distance between the human and the machine, from less than a second to hours. For artificial manipulation, even delays of one second can result in adverse events such as increased bleeding from an open incision or increased frustration and eventual disuse of an advanced prosthesis. Taking a cue from biology, autonomous low-level reflexes that detect stimuli and implement a corrective response in robotic hands without a human in the loop could buy time for communication, information processing, and decision-making in human-machine systems. A long-term research objective of the PI is to advance robotic manipulation with grip reflex algorithm primitives, artificial tactile sensors, and generalizable grasp policy algorithms inspired by the human hand. In this project, she will focus on understanding what drives low-level reactive grip responses, how human-machine performance can benefit from the implementation of similar autonomous primitives, and what grasp policies can be learned by a robotic hand. Contributions of this work will include characterization of the reactive grip responses in human hands, development of human-inspired grip reflex algorithm primitives and tactile sensors for robotic hands, and development of learning algorithms that autonomously extract general grasp policies for robotic hands. Research outcomes will enhance our fundamental understanding of grasp primitives in human hands that provide a foundation for dexterous manipulation, and improve the functionality of robotic hands through grip reflex algorithm primitives and learning algorithms that extract grasp policies. Broader Impacts: This research will transform artificial manipulation by enabling robotic grasp with dynamic control of adduction/abduction degrees-of-freedom and use of biomimetic tactile sensors, thereby revolutionizing robotic manipulators intended for unstructured, access-limited, or unsafe environments (including space, underwater, military, rescue, surgery, assistive, rehabilitative, and prosthetic) that require robustness in the face of uncertainty, control delays, or limited information flow at the human-machine interface. In conjunction with her research the PI will work to engage students at an early age in the exploration of the rich field of robotics. To that end, she will develop hands-on instructional modules for teaching elementary and middle school students about robotics using low-cost materials and deploy them locally for the benefit of students under-represented in science, technology, engineering, and mathematics fields. She will also develop an interactive exhibit for a science museum on robotic hands deploy it locally for the benefit of school-aged children and the general public in the metropolitan Phoenix area.