A machine learning approach to quantify individual gait responses to ankle exoskeletons
Megan R Ebers1, Michael C Rosenberg2, J Nathan Kutz3
1Department of Mechanical Engineering, University of Washington, Seattle, WA, 98195, USA.
Journal of Biomechanics
|July 5, 2023
Summary
A new discrepancy modeling framework quantifies individual exoskeleton responses by analyzing gait changes. This approach helps personalize exoskeleton use for rehabilitation by modeling user-specific reactions.
Area of Science:
- Biomechanics
- Robotics
- Neuroscience
Background:
- Predicting individual responses to exoskeletons is challenging due to a lack of theoretical frameworks for quantifying heterogeneous reactions.
- Understanding user-specific gait changes is crucial for optimizing exoskeleton interventions in rehabilitation.
Purpose of the Study:
- To develop and validate a neural network-based discrepancy modeling framework for quantifying complex gait responses to passive ankle exoskeletons.
- To assess the framework's ability to capture individual exoskeleton responses without prior knowledge of physiological structure or motor control.
Main Methods:
- Utilized neural networks to create models for nominal gait, exoskeleton (Exo) gait, and the discrepancy (response) between them.
- Compared the variance explained by an augmented model (Nominal + Discrepancy) against the Exo gait model and the Nominal model alone.
Main Results:
- The augmented discrepancy model significantly improved the prediction of Exo gait variance compared to the Nominal model for both kinematics and electromyography.
- Median R-squared values for kinematics ranged from 0.928-0.963 and for electromyography from 0.665-0.788, indicating successful quantification of responses.
- Unexplained variance suggests a need for additional measurement modalities or improved resolution for comprehensive response characterization.
Conclusions:
- Discrepancy modeling offers a novel approach to quantify individual exoskeleton responses, enabling personalized rehabilitation strategies.
- The framework successfully models gait changes but highlights the need for enhanced data acquisition for complete response capture.
- This method can accelerate the discovery of individual-specific mechanisms driving exoskeleton responses, paving the way for tailored interventions.
