Doctoral Dissertation Defense – Robert L. McGrath

BME PhD Candidate Robert Levin McGrath will be defending his dissertation:


Robot-Aided Training of Propulsion During Walking Using Pulses of Torque at the Hip and Knee Joints


  • Location: STAR HSC
  • Room #: 113
  • Zoom Link:
  • Password: robot
  • Date: May 11th, 2022
  • Time: 10:30 AM – 12:30 PM, EST
  • Committee: Fabrizio Sergi (Chair), Ryan Zurakowski, Jill Higginson, Ioannis Poulakakis, Darcy Reisman


Following a stroke, survivors often experience hemiparesis which leads to gait impairment often characterized by a reduced gait speed, due to difficulty in generating forward propulsion during walking.  Propulsion is determined by two components: 1) the plantar-flexor moment generated about the ankle and 2) the posture of the trailing limb at push-off; known as trailing limb angle.  Robot-based approaches are attractive for post-stroke rehabilitation due to the reduced physical burden, greater consistency and repeatability, and quantification of lower extremity function.  However, robot therapies and their associated controllers have yet to exceed the efficacy of physical therapy, likely because they do not target a functional gait mechanism such as propulsion.  

The main goal of this dissertation is to develop and test exoskeleton-based training methods to modulate propulsion biomechanics.  We first conducted a study in which we exposed healthy control participants to a factorial modulation of stride length and gait speed and utilize inverse dynamics to estimate the lower extremity joint moments in the sagittal plane.  We observed that stride length (correlated with trailing limb angle) is associated with specific changes in torques of the hip and knee joints during stance.  Informed by the results of this analysis, we designed a set of interaction modes targeting effects in propulsion biomechanics, based on the application of brief pulses of torque at the hip and knee joints during the stance phase of gait.  We tested the effects of these pulses of torque in three different experiments.  In one experiment, pulses of torque were applied in single strides; in a second experiment, pulses were applied in consecutive strides, and in a third experiment, pulses of torque were applied in consecutive strides under user-driven treadmill control.  The hip extension effects during fixed speed training consisted of early stance extension and late stance flexion torques exhibiting increases while a reversal in these torque directions decreased effects during application without strong after-effects, where in user-driven treadmill training, pulse application effects were less prominent but after-effects were large and positive.  For propulsive impulse, fixed treadmill training with late flexion torques exhibited increases and early stance flexion torque exhibited positive after-effects; where user-driven treadmill training only exhibited strong positive effects for early stance extension torques.  

In robotic exoskeleton force-feedback controllers, used as baseline zero-force controllers in training paradigms such as those we have performed in the first Aim, adequate transparency – the minimization of human-robot interaction forces – is an important quality.  Many activities pertaining to rehabilitative robotics, for example, reaching or walking, are cyclical in nature and therefore predictable.  This predictability in the control signal can be exploited if the cyclical signal is continuous and of a finite period which would allow for learning and compensation upon the next cycle, such as in a repetitive controller.  

As such, in the second Aim of this dissertation, we explore the use of repetitive control (RC), for the first time in direct force-feedback control, to reduce the interaction force experienced by a participant during transparent control.  We began with the development, modeling, and application of three forms of repetitive control to a bench-top linear platform to demonstrate the viability of enhancing force-control of human-robot interaction in a quasi-periodic task.  Next, we developed a repetitive controller suitable for application on the direct torque-feedback controlled knee joint of the ALEX II exoskeleton platform, and developed a novel formulation that adjusts the RC period in response to human input.  We found that the two RC formulations exhibited improvement in transparency at the knee joint compared to a standard non-learning controller, but that improvements were at the expense of the hip joint.  To gain insight on the underlying causes of coupled stability and performance of RC applied to multi-joint systems, we developed and applied single-input single-output (SISO) RCs to a dynamic model of the ALEX II exoskeleton.  Via simulation, we discovered that a multi-joint system does not exhibit stability with the the application of SISO RCs, further highlighting the effects of coupled joints on controller performance.