Skip to content

LocoHub

Researchers collect locomotion datasets with different sampling schemes, variable names, and metadata conventions, which makes cross-study analysis slow and error-prone. Manually processing each dataset typically takes hours per person and often requires extra post-processing to repair data issues and align conventions, and folks without a biomechanics background often lack the knowledge to even start that cleanup. LocoHub standardizes these independent datasets into a shared schema, percent-normalized gait cycles, and validated metadata so you can compare studies, reproduce analyses, and build new models without reformatting everything from scratch. Standardized biomechanical datasets and simple tools to load, filter, and analyze them in Python and MATLAB.

What the data looks like

Each row is one point in a percent‑normalized gait cycle (typically 150 samples from 0–100%). Columns include subject/task metadata and standardized biomechanical variables.

subject subject_metadata task task_id task_info step phase_ipsi knee_flexion_angle_ipsi_rad hip_flexion_moment_ipsi_Nm_kg
UM21_AB01 age:25,sex:M,height_m:1.75 level_walking level_walking_normal speed_m_s:1.2,incline_deg:0 1 0.00 0.524 0.85
UM21_AB01 age:25,sex:M,height_m:1.75 level_walking level_walking_normal speed_m_s:1.2,incline_deg:0 1 0.67 0.531 0.82
UM21_AB01 age:25,sex:M,height_m:1.75 level_walking level_walking_normal speed_m_s:1.2,incline_deg:0 1 1.33 0.559 0.81
UM21_AB01 age:25,sex:M,height_m:1.75 level_walking level_walking_normal speed_m_s:1.2,incline_deg:0 1 2.00 0.576 0.80
… … … … … … … … …
UM21_AB01 age:25,sex:M,height_m:1.75 level_walking level_walking_normal speed_m_s:1.2,incline_deg:0 1 99.33 0.507 0.80
  • Metadata columns: subject, optional subject_metadata, task, task_id, task_info, step, and phase_ipsi.
  • Variable columns follow the naming convention joint_motion_side_unit (e.g., knee_flexion_angle_ipsi_rad, hip_flexion_moment_ipsi_Nm_kg).

More details and definitions

Quickstart

import pandas as pd

# Load phase-indexed parquet directly
df = pd.read_parquet('umich_2021_phase.parquet')

# Filter to a subject + task of interest
subset = df[(df['task'] == 'level_walking') & (df['subject'] == 'UM21_AB01')]

# Access normalized phase and a variable
phase = subset['phase_ipsi'].to_numpy()
knee = subset['knee_flexion_angle_ipsi_rad'].to_numpy()

% Load phase-indexed parquet directly (R2021b+)
T = parquetread('umich_2021_phase.parquet');

% Filter to a subject + task of interest
subset = T(T.task == "level_walking" & T.subject == "UM21_AB01", :);

% Access normalized phase and a variable
phase = subset.phase_ipsi;
knee  = subset.knee_flexion_angle_ipsi_rad;

from locohub import LocomotionData

data = LocomotionData('umich_2021_phase.parquet')
subset = data.filter(task='level_walking', subjects=['UM21_AB01'])
cycles, features = subset.get_cycles('UM21_AB01', 'level_walking')

addpath('user_libs/matlab');
loco = LocomotionData('umich_2021_phase.parquet');
level = loco.filterTask('level_walking').filterSubject('UM21_AB01');
[cycles, features] = level.getCycles('UM21_AB01', 'level_walking');

Download Datasets

Dataset Tasks Documentation Clean Dataset Full Dataset
Gtech 2021 Level Walking, Stair Ascent, Stair Descent, Transition Docs Download Download
Umich 2021 Decline Walking, Incline Walking, Level Walking, Run, Sit To Stand, Stair Ascent, Stair Descent, Stand To Sit, Transition Docs Download Download

More details and validation reports: Datasets Reference.

Learn and Contribute

Funding

This work was supported by the National Institute of Biomedical Imaging and Bioengineering of the NIH under Award Number R01EB031166. The content is solely the responsibility of the authors and does not necessarily represent the official views of the NIH.