A machine learning approach to quantify individual gait responses to ankle exoskeletons
Megan R Ebers1, Michael C Rosenberg1,2, J Nathan Kutz3
1Department of Mechanical Engineering, University of Washington, Seattle, WA, 98195, USA.
Biorxiv : the Preprint Server for Biology
|January 30, 2023
Summary
A new neural network framework quantifies individual exoskeleton responses by modeling gait discrepancies. This approach improves prediction accuracy for exoskeleton gait, paving the way for personalized rehabilitation strategies.
Area of Science:
- Biomechanics and Robotics
- Human-Computer Interaction
- Machine Learning in Healthcare
Background:
- Characterizing heterogeneous responses to exoskeleton interventions lacks a robust theoretical framework.
- Predicting individual exoskeleton responses and identifying necessary data remains challenging.
Approach:
- Leveraged a neural network-based discrepancy modeling framework to quantify gait changes in response to passive ankle exoskeletons.
- Developed models for nominal gait, exoskeleton gait, and the discrepancy (response) between them.
- Validated the augmented model's ability to capture exoskeleton responses by comparing variance explained in exoskeleton gait data.
Key Points:
- Discrepancy modeling successfully quantified individual exoskeleton responses without prior physiological or motor control knowledge.
- The augmented model significantly increased variance explained in exoskeleton gait kinematics and electromyography compared to the nominal model (p < 0.042).
- Additional measurement modalities or improved resolution are needed for comprehensive characterization of exoskeleton gait due to unexplained variance.
Conclusions:
- Discrepancy modeling provides a novel framework for understanding individual exoskeleton responses.
- This approach can accelerate the discovery of personalized mechanisms driving exoskeleton effectiveness.
- Enables the development of tailored rehabilitation strategies using exoskeleton technology.
