There is a growing demand for functional rehabilitation orthotics that can effectively assist in patient recovery from motor impairments after stroke. The hand in particular is a complex system that has proven difficult to mimic with current exoskeleton technologies. This paper presents data-driven design parameters to increase the functionality and improve the assistance profile of the ArmAssist-2.0 hand module. Improvements from the previous model include adjustable linkages to fulfill the largest population of users, new joint locations to more accurately represent biomechanics of the hand, and a more impairment-appropriate torque profile to assist in hand opening, adjustable through interchangeable springs. In most passive hand orthoses, assistance forces tend to decrease as the hand and thumb extend, opposite the needs of a typical patient hand. This project utilizes a variable spring moment arm about the revolute axes to match common patient impairment more accurately. The revised assistance profile for the hand maintains a nearly linear relationship. Results conclude that the final assembled device fits comfortably in the hand with noticeable improvements in joint locations, adjustability, and the force profile for the metacarpophalangeal (MCP) joint. An issue arises with the extension of the proximal interphalangeal (PIP) joint due to the nature of rapidly changing moment arms and multiple springs in series. The issue and possible solutions are discussed.

Robotic devices are a promising and dynamic tool in the realm of post-stroke rehabilitation. Researchers are still investigating how the use of robots affects motor learning and what design characteristics best encourage recovery. We present a parallel-actuated, end-effector robot designed to provide spatial assistance for upper-limb therapy while exhibiting low impedance and high backdrivability. A gradient based optimization was performed to find an optimal design that accounted for force isotropy, mechanical advantage, workspace size, and counter-balancing. A beta prototype has been built to these specifications (low impedance and high backdrivability) and has undergone initial controller performance as well as fit and function testing. By fitting a nonlinear model to experimental frequency response data, the apparent mass, viscous friction coefficient, and dynamic dry friction coefficient were determined to be 0.242 kg, 0.114 Ns/m, and 0.894 N respectively. The robot will serve as a testing platform to investigate motor learning and evaluate the efficacy of control schemes for post-stroke movement therapy.