This section contains examples of research from Dr. Shepherd’s PhD and Postdoctoral Work. PhD work was performed at Northwestern University and the Shirley Ryan AbilityLab; Postdoc work was performed at the Georgia Institute of Technology. Collaborators are listed with each project. More projects and project information can be found in the published works listed here.
Design of a Semi-active Variable-Stiffness Prosthetic Ankle (VSPA)
Paper. Collaborators: Elliott Rouse, PhD
Prosthetic feet behave like springs, mimicking the natural function of the ankle-foot complex during walking. However, different tasks (and even different speeds of walking) are better mimicked by springs of different stiffness levels. To enable variation in stiffness across tasks while maintaining much of the light weight and high robustness of common passive devices, we developed a semi-active Variable-Stiffness Prosthetic Ankle. It uses a cantilever spring with a sliding simple support, repositionable by a small DC motor and lead screw. A cam-based transmission also allows for customizable nonlinearity in the Torque-Angle responses, allowing us to better mimic the nonlinear, stiffening behavior of the ankle during gait.
A Passively Dorsiflexing Prosthetic Ankle
(under review) Collaborators: Harrison Bartlett, PhD and Brian Lawson PhD.
When humans walk, they dorsiflex their ankle a couple degrees during swing to increase ground clearance. Typical prosthetic feet, however, do not; this can lead to an increased risk of falls due to trips and scuffs (imagine walking from a tile floor to a carpet while reading your phone–even a 5 mm increase in surface height can dramatically increase trip risk). This prototype ankle, developed by Little Room Innovations (LRI), uses a complex slider-crank action with a soft return spring to dorsiflex 5° when unloaded.
Machine Learning-Based Control of Exoskeletons
Paper. Collaborators: Dean Molinaro, Greg Sawicki PhD, Aaron Young PhD
Standard techniques for controlling exoskeleton torque rely on a running list of previous stride durations, with heel strikes detected by IMUs or force sensors. However, these methods fail to account for accelerations and decelerations of gait speed, as well as natural variability in cadence. In this project, we implement online machine learning-based control of exoskeleton torque by real-time estimating gait phase from a time history of onboard sensor data.
Clinician Preference of Prosthetic Ankle Stiffness
Paper. Collaborators: Elliott Rouse, PhD
In this study, we used a robotic prosthetic ankle and a psychophysics paradigm to simultaneously measure the preferred mechanical behavior (stiffness) of the prosthesis for both the amputee patient, and prosthetists.
Shape Optimization of Running-Specific Prostheses
Paper. Collaborators: Daniel Gunz, Tyler Clites PhD, Christophe Lecomte PhD, Elliott Rouse PhD
In this work, in collaboration with Ossur (Icelandic prosthetics manufacturer), we developed an optimization framework for discovering the shape of a running-specific prosthetic foot from a desired mechanical behavior. The endpoint mechanics can be described by either a set of vertical, horizontal, and angular deflections to a nominal force, or with compliance ellipses indicating the small-force deflection vector as a function of applied force direction. We manufactured custom prosthetic feet to determine the importance of the oft-neglected horizontal deflection induced by vertical ground reaction forces, and found that increasing horizontal deflection increases upstream knee moments.
