Data for: Quantifying human adaptation to a novel split-belt walking condition after broad experience at different belt speeds

Jin, Zijie 1 ; Isa, Jason1; Burden, Samuel1; Ingraham, Kimberly1

Research facility: University of Washington

Published Sep 05, 2025 on Dryad. https://doi.org/10.5061/dryad.rbnzs7hr4

Data files

Sep 05, 2025 version files 21.38 GB

Full_dataset_for_archving.zip

21.38 GB
README.md

3.87 KB

Abstract

This dataset includes raw MATLAB files generated by the Qualisys Track Manager software from motion capture data of 15 non-disabled participants walking on a split-belt treadmill, with only ankle markers. It also contains metabolics measured from the COSMED K5 metabolic cart during treadmill walking. The dataset is associated with the article "Quantifying human adaptation to a novel split-belt walking condition after broad experience at different belt speeds". DOI will be updated shortly after official submission.

Article abstract: Humans can adapt their gait to minimize energy cost when given sufficient exposure to novel energy landscapes. However, it remains unclear whether broad experience in one energy landscape is sufficient to initiate continuous optimization when encountering similar but distinct conditions. In this study, we used visual biofeedback to guide 15 participants to broadly explore walking gaits with different step length asymmetries (SLAs) while walking on a split-belt treadmill with a large split-belt ratio. We subsequently tested how participants adapted their walking patterns during free exploration trials at the large split-belt ratio and at a new, smaller split-belt ratio. Our results showed that during guided exploration, participants were exposed to energetically favorable conditions. However, participants did not self-select walking patterns that minimized their metabolic cost during either free exploration trial. When first exposed to the smaller split-belt ratio, participants immediately adjusted their leg swing distances and belt contact times while maintaining similar step lengths. Throughout this trial, they continued adapting by significantly increasing step lengths on the fast belt. Taken together, our results suggest that participants actively modified their gait strategy when exposed to an energy landscape distinct from the one in which they gained broad experience.

Dataset DOI: 10.5061/dryad.rbnzs7hr4

Description of the data and file structure

Dataset associated with publication "Quantifying human adaptation to a novel split-belt walking condition after broad experience at different belt speeds". https://doi.org/10.1101/2025.08.25.671956

For data collection methods and a detailed list of biomechanical variables analyzed in this dataset, see the associated article link.

Files and variables

The dataset contains two folders. For most purposes (analyzing results, recreating figures from the paper, etc.) it suffices to only look at "processed_data folder", which contains:

A subject_data mat, containing anthropometric information and experiment orders for each participant. The code associated with this dataset will automatically load this.

Processed_data: 15 subfolders (one for each participant) which contain:
1. Raw metabolics data (COSMED) (Excel files) - variables relevant to analyses are VO2 and VCO2, respectively
2. SLA mats processed from QTM (MATLAB mats): these are data processed from raw motion capture output that only includes those variables relevant to analyses included in the publication. The processed MATs mean they have been trimmed from the raw mat files that Qualisys Track Manager produced so that only relevant information is kept. The analysis code associated with this dataset will automatically load these data to generate plots in the publication. Data included:
  - Fx, Fy, Fz : Forceplate readings
  - SL_fast, SL_slow, ST_fast, ST_slow: step lengths and times on fast and slow belt, respectively
    - SLA (step length asymmetry) calculated from step lengths
  - t_HS and t_toeoff, fast and slow: heel strike time and toe off time on fast and slow belt, respectively
  - step_scores: the scores achieved by the participant at every step, collected from real-time (see below)
3. Recorded data from real-time (MATLAB mats). These are the mats saved from participants' real-time biofeedback sessions (see publications for details). Only scores achieved from these mats are used in the publication's analyses - all other analyses used SLA mats processed from QTM. NOTE: the scores are already extracted and included in the processed mats above; these files are only included for reference.
Raw mats from QTM (MATLAB mats): also 15 subfolders, one for each participant. These contain the mats directly exported by QTM from each motion capture session, and include all information, including ankle markers' position, force plate readings, etc., logged by the motion capture software. These are the raw files from which "SLA mats processed from QTM" are generated (script included in the codebase).

Code/software

The dataset is ideally viewed with MATLAB, although Python with packages such as io can readily load them as well. The codes to analyze the data and generate the figures can be found at the following GitHub repo: https://github.com/ingraham-research/Split-belt-paper-2025/tree/main.

The code should be fairly self-explanatory with comments.

Access information

Other publicly accessible locations of the data:

Data was derived from the following sources:

Human subjects data

<p>All participants provided written, informed consent for sharing de-identified data. All participants were designated by number without identifying information. The dataset only contains participants' body mass, leg length, as well as walking motion capture data from positions of two ankle markers and metabolic data from breath analysis.</p>