David Hollinger

Mr. David Hollinger

orcid

Research Focus & Metrics

Academic Career Length

9Years (2018-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
22 March 2023 in IEEE Transactions on Medical Robotics and Bionics
Verified
Journal Article
Closed access
DHDavid Hollinger
David Hollinger
MSMark Schall
Mark Schall
HCHoward Chen
Howard Chen
SBSarah Bass
Sarah Bass
MZMichael Zabala
Michael Zabala
Machine learning has seen a rapid increase in applications that harness wearable signals to improve human mobility. Previous work has used machine learning predictors as a means of continuously estimating locomotor intent. Although previous studies perform gait phase classification and joint-level angular prediction, there are currently no studies that compare joint-level prediction performance at various phases of gait. As such, the purpose of this offline study was to analyze how machine learning and deep learning models perform at predicting future joint angles during various phases of gait. EMG, IMU, and joint kinematics collected during level-ground walking from thirty participants and data was separated into six distinct gait phases. Random forest, long short-term memory (LSTM), and bidirectional LSTM was used to predict lower-limb joint angles during the phases of gait. Results indicate that bidirectional LSTM is the most robust performer across the gait cycle, with a mean prediction RMSE of 1.42-5.71 degrees. Our study shows how deep learning methods, such as bidirectional LSTM, can accurately estimate joint angles throughout the gait cycle. Furthermore, we propose future work of deploying models which accurately predict future joint angles throughout the gait cycle for users to sufficiently operate a wearable exoskeleton during locomotion.

20 May 2020 in PLOS ONE
Verified
Journal Article
Open access
PDF file
Raviraj NatarajRaviraj Nataraj
Raviraj Nataraj
DHDavid Hollinger
David Hollinger
MLMingxiao Liu
Mingxiao Liu
ASAniket Shah
Aniket Shah
This study investigated the generalized effects of positive feedback (PF) versus negative feedback (NF) during training on performance and sense of agency for a reach-to-touch task with a virtual hand. Virtual reality (VR) is increasingly employed for rehabilitation after neuromuscular traumas such as stroke and spinal cord injury. However, VR methods still need to be optimized for greater effectiveness and engagement to increase rates of clinical retention. In this study, we observed that training with disproportionate PF subsequently produced greater reaching performance (minimizing path length) and greater agency (perception of control) than with disproportionate NF. During PF training, there was also progressive increase in agency, but conversely a decrease in performance. Thus, the increase in performance after training may not be due to positively bolstered learning, but rather priming higher confidence reflected in greater agency. Agency was positively measured as compression in perceived time-intervals between the action of touch to a sound consequence, as standard with intentional binding paradigms. Positive feedback desirably increased agency (~180 msec) and reduced path length (1.8 cm) compared to negative feedback, which itself showed insignificant, or neutral, effects. Future investigations into optimizing virtual reality paradigms for neuromotor rehabilitation should consider agency as a driving factor for performance. These studies may serve to optimize how feedback is better presented with performance results for complex motor learning. Investigators should also ponder how personal characteristics, both cognitive and physical, may further affect sensitivity to feedback and the rate of neuromotor rehabilitation.

5 June 2024 in Sensors
Verified
Journal Article
Open access
PDF file
DHDavid Hollinger
David Hollinger
MSMark Schall
Mark Schall
HCHoward Chen
Howard Chen
MZMichael Zabala
Michael Zabala
The use of wearable sensors, such as inertial measurement units (IMUs), and machine learning for human intent recognition in health-related areas has grown considerably. However, there is limited research exploring how IMU quantity and placement affect human movement intent prediction (HMIP) at the joint level. The objective of this study was to analyze various combinations of IMU input signals to maximize the machine learning prediction accuracy for multiple simple movements. We trained a Random Forest algorithm to predict future joint angles across these movements using various sensor features. We hypothesized that joint angle prediction accuracy would increase with the addition of IMUs attached to adjacent body segments and that non-adjacent IMUs would not increase the prediction accuracy. The results indicated that the addition of adjacent IMUs to current joint angle inputs did not significantly increase the prediction accuracy (RMSE of 1.92° vs. 3.32° at the ankle, 8.78° vs. 12.54° at the knee, and 5.48° vs. 9.67° at the hip). Additionally, including non-adjacent IMUs did not increase the prediction accuracy (RMSE of 5.35° vs. 5.55° at the ankle, 20.29° vs. 20.71° at the knee, and 14.86° vs. 13.55° at the hip). These results demonstrated how future joint angle prediction during simple movements did not improve with the addition of IMUs alongside current joint angle inputs.

3 June 2024 in IEEE Transactions on Medical Robotics and Bionics
Verified
Journal Article
Closed access
RPRyan Pollard
Ryan Pollard
DHDavid Hollinger
David Hollinger
INIván E. Nail-Ulloa
Iván E. Nail-Ulloa
MZMichael E. Zabala
Michael E. Zabala
Elevated runtimes of machine learning algorithms and neural networks make their inclusion in near-future joint angle estimation difficult. The purpose of this study was to develop simple, analytical models that prioritize historical joint kinematics when estimating near-future joint angles. Five kinematically-informed and extrapolation-based methods were developed for joint angle estimation at three near-future estimation horizons: tpred = 50 ms, 75 ms, and 100 ms. The estimation error and required runtimes of each prediction algorithm were evaluated on the sagittal-plane ankle angles of 24 individual subjects who performed three level-ground walking trials. Results showed that the kinematically-informed models had significantly faster estimation runtimes than Random Forest (RF) machine learning models trained and tested on identical datasets (kinematic models: trun < 0.62 ms, RF models: trun > 8.19 ms for all estimation horizons). The RF models exhibited significantly lower prediction errors than the kinematic models for estimation horizons of tpred = 75 ms and 100 ms, but no significance was found between the top-performing kinematic model and RF models for a tpred = 50 ms. These results indicate that a kinematically-informed approach to joint angle estimation can serve as a simple alternative to complex machine learning models for very near-future applications (tpred ≤ 50 ms) while serving as a comparison baseline for more distant estimation horizons (tpred ≥ 75 ms).

11 April 2020
Verified
Book Chapter
Closed access
PDF file
Raviraj NatarajRaviraj Nataraj
Raviraj Nataraj
SSSean Sanford
Sean Sanford
MLMingxiao Liu
Mingxiao Liu
KWKevin Walsh
Kevin Walsh
SWSamuel Wilder
Samuel Wilder
DHDavid Hollinger
David Hollinger

1 October 2023 in ASME Letters in Dynamic Systems and Control
Verified
Journal Article
Closed access
PDF file
ABAvinash Baskaran
Avinash Baskaran
DHDavid Hollinger
David Hollinger
RHRhet O. Hailey
Rhet O. Hailey
MZMichael Zabala
Michael Zabala
CRChad G Rose
Chad G Rose
Robotic exoskeletons for the hand are being explored to improve health, safety, and physical performance. However, much research effort is needed to establish reliable models of human behavior for effective human–robot interaction control. In this work, surface electromyography is used to measure and model muscle activity of healthy participants performing quasi-isometric and dynamic hand exercises. Non-negative matrix tri-factorization is used to extract hidden neuromuscular parameters encoded in spatial and temporal muscle synergies, which are used to estimate probabilistic linear models of intent, effort, and fatigue. This paper thereby presents steps toward reliable modeling of nonlinear time-varying hand neuromuscular dynamics for intuitive and robust human–robot interaction.