Reshaping Motor Learning in High-Dimensional Tasks via Soft Robotic Physical Interactions

Michigan State University, East Lansing MI

Investigators

Abstract

Movement impairments after neurological injury such as stroke are one of the leading causes of long term disability in the United States. Impairments
 of the upper extremity, including hand and finger
function, are extremely common, often resulting in difficulties performing
daily activities. 
Neurorehabilitation often requires learning coordination patterns that involve the complex motion of a large number of joints in the body. Because common approaches such as instructions and visual feedback are largely ineffective for guiding such complex motor learning, this project aims to understand and facilitate the learning of complex coordination patterns by applying forces using robots. The project will lead to development of a soft-robotic glove and associated algorithms for understanding and facilitating motor learning. Understanding how to guide motor learning in complex tasks has a wide range of applications in stroke rehabilitation, skill-training for athletes, collaborative human-robot manipulation, and robot-assisted surgery among others. Advances made in this project will thus have significant societal impact in these areas. The multi-disciplinary research is integrated with outreach and educational activities that aim to broaden the participation of underrepresented groups and to increase involvement of middle and high school students in engineering research. Drawing on unique and complementary expertise of the team in motor learning, computational modeling, robotics, and control, this research project will result in a rigorous, systematic framework for modeling and facilitating bidirectional human-robot learning, via physical interactions, for tasks involving a large number of degrees of freedoms. The project goal will be achieved through four integrated research thrusts: (1) developing a soft, compact, and sensor-rich robotic glove that can apply desired assistance or resistance (emulating impairment) to individual finger joints; (2) developing dynamic model decomposition (DMD)-based data-driven models that capture the learning dynamics and the role of robotic assistance/resistance; (3) exploiting the DMD-based models and model predictive control theory to design the desired robotic assistance as well as the sequence of target locations, to facilitate rapid learning; and (4) evaluating the research approach with extensive experiments, by examining the evolution of the task performance and the underlying coordination patterns of hand joints. The motor learning will be evaluated on in a motion coordination task in which motion of the joints of the hand will be mapped to the up, down, right and left motion of a cursor on a screen. 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.