Characterizing associations between simulated muscle coordination constraints and data-driven representations of human walking dynamics
Paul, Christian
Paul, Christian
Citations
Altmetric:
License
License
Abstract
Personalizing gait rehabilitation is critical for effectively restoring walking function following a neurological injury like a stroke, but remains challenging. Individual differences in neural and biomechanical (i.e., neuromechanical) impairments likely underlie differential changes in gait patterns (e.g., joint kinematics) during treatment, but those differences cannot be identified from gait kinematics alone. Neuromechanical impairments, such as altered muscle coordination in stroke survivors, shape how a person's gait evolves over time and in response to treatment (i.e., a person's gait dynamics). Therefore, individual differences in impairment may be better characterized by differences in gait dynamics than by differences in kinematics. The state-of-the-art approach to encoding dynamics is through musculoskeletal models, which are physiologically detailed reconstructions of the musculoskeletal system. However, these models require assumptions about an individual's physiology, which are hard to validate for a specific individual. One solution to this challenge is to encode neuromechanical dynamics using machine learning, which discovers individual-specific gait dynamics from timeseries joint kinematics, without explicit assumptions about an individual's physiology. However, comparing these gait dynamics between individuals is challenging. Discrepancy modeling enables inter-individual comparisons of dynamics by quantifying differences in dynamics based on vector fields. Here we are determining whether discrepancy modeling of machine-learning-based gait dynamics reveal differences in simulated muscle coordination impairments. While we previously found that machine learning based signatures of gait neuromechanics captured individual differences in post-stroke gait neuromechanics, we show here that these signatures are sensitive to kinematics. We developed and validated a Hankel Alternative View of Koopman (HAVOK) model to encode gait dynamics from simulated kinematics. We will analyze whether HAVOK-based discrepancy models capture individual differences in gait dynamics in simulations with impaired muscle coordination. HAVOK-based discrepancies that capture gait dynamics may, therefore, be useful in encoding individual differences in simulated muscle coordination during walking, potentially informing rehabilitation personalization.
Description
Date
1/1/2026
Student Status
Junior (Graduating in 2027)
Journal Title
Journal ISSN
Volume Title
Type of presentation
Poster
Collections
Research Projects
Organizational Units
Journal Issue
Citation
DOI
Advisor(s)
Department
Program/Major
Computer Science
College/School
College of Engineering and Mathematical Sciences
Organization
Research Category
Engineering