This Faculty Early Career Development Program (CAREER) grant will develop a novel game-theoretic framework addressing the emerging challenge of joint intent inference during physical human/machine interaction. Physical human-machine interaction has become ubiquitous in collaborative manufacturing, robot-aided rehabilitation, and robotic surgery. However, most existing robots lack high-level intelligence to reason about, adapt to, and potentially shape human actions within the context of collaborative tasks. This interdisciplinary project merges game theory, optimal control methods, and ideas gleaned from cognitive hierarchy theory to address the joint inference problem within the exemplar application of a powered knee exoskeleton used for gait rehabilitation. This project will promote the progress of science and advance the national health by developing a novel control framework that will enable an embodied machine intelligence to understand noisy human motor actions, thereby facilitating human learning of motor tasks. In doing so, this project will improve user experience and the productivity of future human-robot teams. The impact of this project will be broadened by developing new curriculum on human-robot interaction, through joint workshops with healthcare partners, and by offering summer internships for Science, Technology, Engineering, and Math (STEM) teachers.
This project addresses significant challenges pertaining to joint intent inference during human-machine physical interaction. The application is that of a powered knee exoskeleton used for gait rehabilitation. The project pursues three specific research activities. The first creates a model of the human's cognitive state when interacting physically with a robot. The model explicitly considers human-machine interaction dynamics, mutual learning, and multi-level intent inference. The second formulates a motion planning and control problem, which will allow the robot both to signal its intents and to facilitate human motor learning based on the identified cognitive and biomechanical models of the human. The third will quantify human performance variability and use it to improve intent inference and signaling performance. Human subject experiments will provide the data needed to build the model of human cognitive state (including intent), to demonstrate improved human-robot team performance over existing robot controllers, and to help understand fundamental mechanisms contributing to effective physical human-robot interaction. A key innovation of this project is the creation of a task-agnostic framework that leverages models of human cognitive and motor dynamics such that an intelligent machine can dynamically adjust its behavior to simultaneously facilitate human learning and provide physical assistance when needed.
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.
