Data-driven spectral analysis for coordinative structures in periodic human locomotion

Keisuke Fujii, Naoya Takeishi, Benio Kibushi, Motoki Kouzaki, Yoshinobu KawaharaScientific 2019 · DOI

TL;DR

This paper introduces a novel data-driven spectral analysis method, Hankel Dynamic Mode Decomposition (Hankel DMD), to uncover the underlying dynamical properties of coordinative structures in biological periodic movements like human walking. By treating these movements as nonlinear limit-cycle oscillations, the method extracts not only the spatial patterns (coordinative structures) but also their temporal dynamics (frequency and phase), offering a more complete understanding than traditional dimensionality reduction techniques.

What Problem Does This Paper Address?

Understanding how complex biological systems, with their numerous degrees of freedom (like the many joints in the human body), achieve coordinated and flexible movements is a long-standing challenge. Traditional methods for identifying "coordinative structures" – simplified representations of these complex movements – often rely on statistical dimensionality reduction (e.g., PCA). While these methods can identify spatial patterns, they typically fail to capture the crucial dynamical properties, such as the frequency and phase of these movements. This limits our ability to understand how these structures evolve over time and adapt to different conditions, like walking at various speeds. The paper aims to bridge this gap by developing a method that extracts both the spatial and temporal dynamics of coordinative structures from observational data.

Methods

The core of the paper's contribution lies in the application of operator-theoretic spectral analysis, specifically a data-driven approach called Dynamic Mode Decomposition (DMD), and its variant, Hankel DMD.

  • Koopman Operator: The authors frame the problem by considering the system's evolution through a Koopman operator, which acts linearly on observable functions, even for nonlinear systems. Spectral analysis of this operator can reveal the system's dynamical properties.
  • Dynamic Mode Decomposition (DMD): DMD is a popular algorithm for approximating the Koopman operator's spectral decomposition from data. It decomposes a time series into spatial modes (coordinative structures) and their associated temporal dynamics (eigenvalues representing frequency and growth/decay rates).
  • Hankel DMD: A key innovation is the use of Hankel DMD. Standard DMD can struggle with limited data or when the underlying dynamics are not fully captured by the observed state. Hankel DMD addresses this by using delay embedding to create Hankel matrices from the input data. This augmentation allows for a more robust estimation of Koopman eigenvalues and eigenfunctions, which are crucial for understanding the phase dynamics of periodic systems. The paper specifically explores two types: row-type (better for phase estimation) and column-type.
  • Phase Reduction: The authors leverage Hankel DMD to perform a form of phase reduction, approximating complex limit-cycle oscillators with simpler phase models, which is particularly useful for understanding biological periodic phenomena.

Datasets

The study utilizes two main types of data:

  • Human Locomotion Data: Kinematic data (segmental angles) from healthy male participants walking on a treadmill at 14 different speeds (2.0 to 8.0 km/h, plus a preferred speed). This data was collected using a 3D optical motion capture system at 100 Hz.
  • Simulation Data:
    • Double Pendulum: A system with known analytical solutions and no redundancy, used for validation.
  • Walking Model: A multi-link bipedal model with explicit governing equations but less redundancy than actual human locomotion, also used for validation.

Key Insights and Ideas

  • Dynamics Matter: Traditional methods like SVD capture spatial patterns but miss the crucial temporal dynamics (frequency, phase) that govern how these patterns evolve. Hankel DMD bridges this gap.
  • Harmonics of Gait: Human walking isn't just at the fundamental gait frequency; it exhibits significant contributions from its harmonics. Hankel DMD effectively extracts these harmonic frequencies, providing a richer spectral representation.
  • Speed-Dependent Phase Evolution: While the overall coordinative structures (spatial patterns) might be similar across different walking speeds, the phase of these structures evolves differently. This speed-dependent phase behavior is a key finding, suggesting adaptive control mechanisms.
  • Coordinative Structures are Dynamic: The paper demonstrates that coordinative structures are not static entities but are intrinsically linked to the underlying dynamics of the system, evolving in frequency and phase.
  • Data-Driven Insight into Complex Systems: The approach offers a powerful way to analyze biological systems with unknown and redundant dynamics, where explicit mathematical models are difficult to derive.

Key Results

  • Harmonic Extraction: Both row-type and column-type Hankel DMD successfully extracted five harmonic frequencies of the gait frequency, outperforming other DMD variants and Fourier spectra in terms of clarity and distinctness.
  • Speed-Dependent Phase: Visualizations of Koopman eigenfunctions (phases) showed that while the overall trajectory shapes were similar across speeds, the phase evolution was heterogeneous and speed-dependent, particularly for the fundamental gait frequency.
  • Reconstruction Performance: Column-type Hankel DMD showed the best reconstruction performance among the DMD methods tested, with significantly lower reconstruction errors and higher VAF compared to exact and companion-matrix DMD.
  • Validation: The method's effectiveness was confirmed using simulation data from a double pendulum and a walking model, showing consistent results regardless of system redundancy.
  • Foot Angle Dominance: Analysis of intersegmental weights (DMD modes) indicated that foot angles robustly contributed to global dynamics at harmonic frequencies more than thigh and shank angles, especially at higher walking speeds.

Limitations and Future Work

  • Mechanical Interpretation: While the method extracts frequencies and phases, a strict mechanical interpretation of higher harmonic frequencies is still needed.
  • Ideal Cyclic Motion Assumption: The current approach assumes underlying limit-cycle dynamics. Future work should adapt to non-ideal cyclic motions, rhythm shifts, and perturbations (e.g., phase resetting).
  • Participant Variation: The study focused on speed variations. Further research could explore inter-participant variability and different locomotion types.
  • Validation for Unknown Dynamics: A general quantitative validation method for systems with unknown governing equations is still required to definitively prove the extraction of "true" dynamical properties.
  • Control Applications: The authors suggest future applications in control problems using forward simulations, exploring global entrainment and reflex actions.

Summary updated

Abstract

Living organisms dynamically and flexibly operate a great number of components. As one of such redundant control mechanisms, low-dimensional coordinative structures among multiple components have been investigated. However, structures extracted from the conventional statistical dimensionality reduction methods do not reflect dynamical properties in principle. Here we regard coordinative structures in biological periodic systems with unknown and redundant dynamics as a nonlinear limit-cycle oscillation, and apply a data-driven operator-theoretic spectral analysis, which obtains dynamical properties of coordinative structures such as frequency and phase from the estimated eigenvalues and eigenfunctions of a composition operator. Using segmental angle series during human walking as an example, we first extracted the coordinative structures based on dynamics; e.g. the speed-independent coordinative structures in the harmonics of gait frequency. Second, we discovered the speed-dependent time-evolving behaviours of the phase by estimating the eigenfunctions via our approach on the conventional low-dimensional structures. We also verified our approach using the double pendulum and walking model simulation data. Our results of locomotion analysis suggest that our approach can be useful to analyse biological periodic phenomena from the perspective of nonlinear dynamical systems.
BibTeX
@inproceedings{fujii2019data,
  author = {Keisuke Fujii and Naoya Takeishi and Benio Kibushi and Motoki Kouzaki and Yoshinobu Kawahara},
  title = {{Data-driven spectral analysis for coordinative structures in periodic human locomotion}},
  booktitle = {Scientific},
  year = {2019},
  doi = {10.1038/s41598-019-53187-1},
}
Figure