Reference

Concise, complete description of what’s in the standardized data.

Data Formats

Format	File Pattern	Index	Purpose
Phase‑Indexed	*_phase.parquet	150 samples, 0–100%	Cross‑subject comparisons, averaging
Time‑Indexed	*_time.parquet	Seconds	Event detection, raw analysis

Phase-aligned exports apply to cyclic gait data. Event-driven or non-cyclic activities should favor the time-indexed format, optionally providing per-repetition segments through the step column.

Required Columns

Column	Required	Meaning
subject	Yes	Global unique ID (see Subject Naming)
subject_metadata	Optional	Demographics, key:value (e.g., age:25,sex:M,height_m:1.75,weight_kg:70)
task	Yes	Activity category (see Task Names)
task_id	Yes	Specific variant (see Task IDs)
task_info	Yes	Parameter string, key:value (see Task Metadata)
step	Yes	Cycle index within trial
phase_ipsi	Phase only	0–100, phase of ipsilateral limb (heel‑strike to heel‑strike)
time_s	Time only	Seconds from trial start

Task Names, IDs, and Metadata

The canonical list of task families lives in internal/config_management/task_registry.py. Update that module and this section whenever new tasks are introduced so that all tooling shares a single source of truth.

• Task names (task) capture the high-level motion family. Use neutral categories that apply across datasets, such as level_walking, incline_walking, decline_walking, stair_ascent, stair_descent, run, walk_backward, sit_to_stand, stand_to_sit, step_up, step_down, jump, squats, and the catch-all functional_task for non-cyclic or miscellaneous movements. Append population/pathology suffixes only when the entire recording targets that cohort. Use the convention task_family_<pathology> (e.g., level_walking_stroke, run_pd, stair_descent_sci). Keep suffix tokens lowercase snake_case (stroke, pd, sci, tfa, tta, etc.).
• Task IDs (task_id) describe the specific variant within the family, e.g., level, incline_5deg, decline_10deg, stair_ascent, stair_descent, sit_to_stand_short, jump_vertical. The exact vocabulary can differ by dataset, but it must remain stable inside a release.
• Task metadata (task_info) is a comma-separated key:value string carrying numeric parameters and variant tags. Values should prefer SI units and be machine-readable.

Managing Task Families with the CLI

Use python contributor_tools/manage_tasks.py to add, delete, or list canonical task families without editing Python files by hand. Reach for the CLI when you introduce new base families that do not already exist in the registry. Examples:

• python contributor_tools/manage_tasks.py list --category phase
• python contributor_tools/manage_tasks.py add lateral_walk --category phase --description "Side-step gait cycles" --example-id lateral --notes "Segment heel strikes the same way as level_walking."
• python contributor_tools/manage_tasks.py delete demo_task

Pathology or cohort variants follow the task_family_<pathology> convention but do not require separate registry entries. Instead, create dedicated validation ranges for each suffix (e.g., level_walking_stroke) so tooling can read the cohort-specific tolerances. Validators never fall back to the able-bodied family values; use the interactive tuner to seed a new ranges file with the base family as an initializer before adjusting to the cohort norms.

Common task_info keys

• speed_m_s:<float>
• treadmill:<true|false>
• incline_deg:<float> (positive uphill, negative downhill)
• step_height_m:<float> (stairs)
• step_width_m:<float> (stairs)
• surface:<string> (e.g., overground, treadmill)
• footwear:<string> (e.g., barefoot, shoe)
• assistance:<string> (e.g., none, handrail)
• transition_from:<task> / transition_to:<task> (for mode switches)
• turn_direction:<left|right|cw|ccw> (for turning strides)
• variant:<string> (label specific instructions when no numeric parameter exists)

Task Definitions

Task families fall into two groups. Phase-friendly tasks can be normalized to 150-sample strides when consistent events exist (and may also be exported in time if desired). Time-indexed tasks lack repeatable kinematic cycles or are best analyzed as continuous episodes. The tables below summarize each group; detailed notes follow.

Phase-Friendly Families

When segmented, these tasks are normalized to 150 samples per stride with phase_ipsi running 0–100%. Phase output is preferred because it enables stride averaging and template validation; however, a time-indexed export is acceptable when segmentation is unavailable.

Task	Typical behaviors	Example task_id values	Phase definition	Core metadata	Notes
level_walking	Overground/treadmill at steady speed	level, level_fast, level_slow	0% ipsilateral heel strike, 50% contralateral heel strike, 100% next ipsilateral heel strike	speed_m_s, treadmill, surface	Expect near-symmetric sagittal kinematics; large pelvis/trunk offsets flag coordinate issues.
incline_walking	Uphill ramps/treadmills	incline_5deg, incline_10deg	Same as level walking (HS-to-HS)	incline_deg, treadmill, assistance	Record grade and surface; uphill cycles show increased hip/knee flexion and trunk lean.
decline_walking	Downhill ramps/treadmills	decline_5deg, decline_10deg	Same as level walking with downhill HS events	incline_deg (negative), treadmill	Negative grades with delayed toe-off; expect larger eccentric knee moments.
stair_ascent	Ascending standard stairs	stair_ascent, numbered passes	0% ipsilateral contact on current step, 100% ipsilateral contact on next step	step_height_m, step_number, assistance	Document geometry and handrail use; monitor hip/knee angles >90°.
stair_descent	Descending stairs	stair_descent, numbered passes	0% upper-step contact, 100% lower-step contact for ipsilateral foot	step_height_m, step_number, assistance	Same metadata as ascent; eccentric control dominates mid-step.
run	Jogging/running with flight phases	run_2_5_m_s, run_3_0_m_s	0% ipsilateral foot contact, 100% next ipsilateral contact including flight	speed_m_s, treadmill, surface, footwear	Two GRF peaks with flight intervals; watch pelvis rotation extremes.
transition	Gait-to-gait transitions	walk_to_run, stair_to_walk, turn	0% key event of departing gait (e.g., heel strike), 100% first event of target gait	transition_from, transition_to, gait_transition	A single stride may cover two behaviors; store context keys and treat directions separately.
sit_to_stand / stand_to_sit	Chair or box transfers	sit_to_stand, stand_to_sit	Sit→stand exported 0–50% from seated start to stable upright; phase 50% represents standing. Stand→sit exported 50–100% from upright start back to seated contact, sharing the 50% standing convention.	chair_height, armrests, variant	Segment seat-off/seat-on events for ascent/descent cycles when possible.
squats	Loaded or bodyweight squats	squats, squat_down, squat_up	0% upright start, 50% lowest depth, 100% return to upright	weight_kg, variant, cadence	Split descent/ascent for phase use; keep full set in time if reps are irregular.
step_up / step_down	Stair-box repetitions, curbs	step_up, step_down	Step-up: 0% initial foot contact on box, 100% full weight on box. Step-down mirrors with lower surface contact.	height_m, lead_leg, step_number	Treat like short stair runs; ensure cadence is repeatable before phase export.
jump	Hops, vertical/lateral jumps	jump_vertical, jump_lateral, hop_single	0% preparatory contact, 50% takeoff, 100% landing contact of same foot	jump_type, variant, lead_leg	Use takeoff→landing cycles for repetitive sets; single attempts may stay time-indexed.
weighted_walk, dynamic_walk, walk_backward	Walking variants with perturbations	level, variant:<string>	Same as level walking (HS-to-HS) when heel strikes are present	speed_m_s, treadmill, variant	Prefer phase when heel strikes exist; otherwise document perturbation under variant.

Transfers (sit_to_stand, stand_to_sit) intentionally share a 50% phase landmark that corresponds to an upright, standing posture. Sit-to-stand cycles occupy 0–50% phase and end in standing, while stand-to-sit cycles begin from that same standing posture at 50% and end seated at 100%. Use this convention when deriving new datasets so comparative analyses treat the standing posture consistently across ascent and descent.

Time-Indexed (Non-Cyclic) Families

Export these tasks as time-indexed episodes. They lack consistent heel-strike structure, involve perturbations that break cyclic assumptions, or represent single-shot activities. If future datasets introduce segmentation for any of these behaviors, they can graduate to the phase-friendly group.

Task	Typical behaviors	Example activities / task_id	Required metadata	Notes
agility_drill	Multi-directional or cadence-driven drills without steady gait cycles	side_shuffle, tire_run, dynamic_walk_high_knees, turn_and_step_left	variant, direction, cadence_hz, surface	Capture cueing or treadmill settings; use when strides vary too much for phase export.
cutting	Athletic cuts with large heading changes	cutting_left_fast, cutting_right_slow	direction, approach_speed_m_s, turn_angle_deg, surface	Keep entire maneuver in time to preserve path-dependent kinetics.
free_walk_episode	Episodic or exploratory walking sequences	meander, start_stop, obstacle_walk	surface, segment_start_s, segment_end_s, treadmill	Use for trials with frequent pauses/obstacles where heel-strike segmentation is unreliable.
load_handling	Asymmetric lifts, carries, or object exchanges	lift_weight_25lbs_l_c, ball_toss_center	load_kg, hand, pickup_height, dropoff_height, direction	Document object mass and interaction side so bias/torque effects are traceable.
perturbation	External pushes/pulls or cooperative force tasks	push_forward, tug_of_war, twister	direction, magnitude, support, implements	Distinguish voluntary vs. reactive responses via metadata (e.g., initiated_by:experimenter).
balance_pose	Static or quasi-static holds	poses_single_leg, poses_wide_stance, squat_hold	variant, duration_s, support	Useful for calibration and balance assessments; ensure durations cover full hold.
functional_task	Miscellaneous non-cyclic behaviors not covered above	variant:<string>	variant, contextual keys as needed	Use sparingly; prefer one of the specific families when possible.

When new datasets introduce additional behaviors, extend the table with the task family, expected metadata, and note whether improved segmentation would allow promotion to the phase-friendly bucket.

Subject Naming

Format: <DATASET_CODE>_<POPULATION_CODE><SUBJECT_NUMBER> → e.g., DS23_AB05, DS21_TFA03. Keep subject numbers zero-padded to two digits where possible to aid sorting.

Population codes

Code	Population	Recommended task suffix
AB	Able‑bodied	(omit)
TFA	Transfemoral amputee	tfa
TTA	Transtibial amputee	tta
CVA	Stroke	stroke or cva
PD	Parkinson’s	pd
SCI	Spinal cord injury	sci
CP	Cerebral palsy	cp
TKA	Total knee arthroplasty	tka
THA	Total hip arthroplasty	tha
MS	Multiple sclerosis	ms

Subject Metadata

Optional demographics and clinical context stored in subject_metadata as comma‑separated key:value pairs.

Common keys

• age:<int>
• sex:<M|F|Other>
• height_m:<float>
• weight_kg:<float>
• leg_dominance:<left|right|unknown>
• impairment:<string> (e.g., stroke, pd, sci)
• prosthesis_type:<string> (e.g., TFA, TTA)
• prosthesis_side:<ipsi|contra|left|right>
• clinical_scores:<string> (e.g., FuglMeyer:28)
• notes:<string> (short free text)

Variables (What Columns Mean)

Naming pattern: <joint/segment>_<motion>_<measurement>_<side>_<unit>

Key categories

• Kinematics (angles): *_angle_*_rad
• Kinetics (moments/forces): *_moment_*_Nm_kg (mass-normalized)
• Ground reaction forces (GRF): *_grf_*_BW (body-weight-normalized)
• Segment orientations: pelvis_*_angle_rad, trunk_*_angle_rad, thigh_*_angle_*_rad, shank_*_angle_*_rad, foot_*_angle_*_rad

Sides

• ipsi: limb that defines phase (phase_ipsi is 0% at its heel strike)
• contra: opposite limb (≈ 50% phase offset in level walking)

Units

• Angles in radians (*_rad)
• Moments are mass‑normalized (*_Nm_kg)
• GRFs are body‑weight‑normalized (*_BW)

Column Catalog

Complete list of standard column names used in the conversion scripts and examples. Columns are grouped by category; sides are ipsi and contra unless noted.

• Required schema: subject, subject_metadata, task, task_id, task_info, step, phase_ipsi (phase‑indexed) or time_s (time‑indexed)
• Optional schema: phase_contra, dataset, collection_date, processing_date

Kinematics — joint angles (radians)

• hip_flexion_angle_{ipsi,contra}_rad
• hip_adduction_angle_{ipsi,contra}_rad
• knee_flexion_angle_{ipsi,contra}_rad
• ankle_dorsiflexion_angle_{ipsi,contra}_rad

Kinetics — joint moments (Nm/kg)

• hip_flexion_moment_{ipsi,contra}_Nm_kg
• knee_flexion_moment_{ipsi,contra}_Nm_kg
• ankle_dorsiflexion_moment_{ipsi,contra}_Nm_kg

Segment/link orientations (radians)

• Pelvis: pelvis_sagittal_angle_rad, pelvis_frontal_angle_rad, pelvis_transverse_angle_rad
• Trunk: trunk_sagittal_angle_rad, trunk_frontal_angle_rad, trunk_transverse_angle_rad
• Thigh: thigh_sagittal_angle_{ipsi,contra}_rad
• Shank: shank_sagittal_angle_{ipsi,contra}_rad
• Foot: foot_sagittal_angle_{ipsi,contra}_rad

Angular velocities (radians/second)

• hip_flexion_angular_velocity_{ipsi,contra}_rad_s
• knee_flexion_angular_velocity_{ipsi,contra}_rad_s
• ankle_dorsiflexion_angular_velocity_{ipsi,contra}_rad_s

Ground reaction forces (BW)

• Vertical: vertical_grf_{ipsi,contra}_BW
• Anterior–posterior: anterior_grf_{ipsi,contra}_BW
• Medio–lateral: lateral_grf_{ipsi,contra}_BW

Center of pressure (meters)

• cop_anterior_{ipsi,contra}_m
• cop_lateral_{ipsi,contra}_m

Coordinate & Sign Conventions

OpenSim right‑handed frames. Positive rotations follow right‑hand rule.

• Hip flexion (+): thigh forward; Knee flexion (+): heel toward buttocks
• Ankle dorsiflexion (+): toes up; Pelvis sagittal (+): anterior tilt

Row & Phase/Time Semantics

• Each row = one sample within a step/cycle for a subject+task.
• Phase‑indexed: phase_ipsi runs 0→100% in each gait cycle of the ipsilateral limb (150 samples).
• Contra limb is approximately 50% offset in level walking.
• Time‑indexed: time_s increases monotonically; cycles delimited by gait events.

GRF & CoP Variables

• GRF components: vertical_grf_*, anterior_grf_*, lateral_grf_* with suffix _BW.
• Center of pressure: cop_anterior_*_m, cop_lateral_*_m (meters), side‑specific where applicable.

Naming Rules

• Lowercase snake_case; tokens separated by single underscores.
• Use exact unit suffix tokens: rad, Nm_kg, BW, m, s.
• Segment/global (no side): e.g., pelvis_*_angle_rad, trunk_*_angle_rad.
• Side tokens: ipsi or contra only.

Optional Columns & Flags

• phase_contra (percent): contra limb phase, if present.
• cycle_id: optional alias of step.
• dataset: dataset identifier; collection_date, processing_date (ISO strings) if available.
• Quality flags (boolean): e.g., is_reconstructed_<side> for interpolated values.

Canonical phase column is phase_ipsi.

Related

• Datasets overview: ../datasets/
• Validation ranges: ../datasets/validation_ranges.md
• Maintainers: ../maintainers/