OpenGait is an open specification for gait and fall-risk tracking. It defines what to measure, how to compute it, the reference risk bands, and one data format — so a phone, a sensor insole, a wearable, or a device you build in your garage all speak the same language. Implement it freely. Bring your own hardware. Or build your own.
OpenGait doesn't sell you a box. It defines a standard, and gives everyone an on-ramp — from a maker with an Arduino to a hospital with a fleet of insoles.
Make a conformant tracker from off-the-shelf parts and emit the standard format. Two reference paths:
Reference designs + BOM provided to implementers. If your output validates, you're conformant.
Already have a sensor insole, a smartwatch, a research rig, an accelerometer? Write a thin adapter that maps its output to the OpenGait data model. No re-instrumenting.
Run your implementation against our reference test walks (known ground truth). Pass, and you may display the OpenGait Conformant mark and interoperate with any OpenGait consumer.
The core gait observation. Units are fixed; computation methods are documented so two devices produce comparable numbers. Designed to map cleanly onto a FHIR Observation with LOINC coding.
| Field | Unit | Definition |
|---|---|---|
| gait_speed | m/s | Self-selected walking speed over a steady-state segment. The headline metric — the 6th vital sign. |
| cadence | steps/min | Steps per minute during the segment. |
| stride_length | m | Distance between successive same-foot contacts. |
| symmetry | 0–100 | Left/right balance of stride and timing; 100 = symmetric. |
| double_support | % cycle | Share of the gait cycle with both feet down — rises with instability. |
| variability | CoV % | Stride-to-stride variability (coefficient of variation) — a key fall predictor. |
| capture_method | enum | camera · imu_insole · pressure_insole · wearable · timed_walk |
| confidence | 0–1 | Estimator's confidence in the reading; lets consumers weight or discard. |
Implementation note — feed estimators triaxial (x, y, z) accelerometry, not collapsed magnitude. The open Inertia-1 study (2026) finds that direction of motion carries transferable structure a single magnitude channel throws away — triaxial input measurably outperforms it across activity, gait, and disease tasks. OpenGait's imu_insole and wearable capture methods should preserve all three axes upstream of the metrics above.
OpenGait publishes a transparent reference fall-risk band set, drawn from the gait-speed literature. Anyone can implement the same risk read. (Production-grade trajectory models can be built on top — see §05.)
The model that reads raw motion is commoditizing too. In July 2026, UCLA, Duke, and Johns Hopkins released Inertia-1, a fully open wearable-motion foundation model trained on 18.2M hours from 115,000+ people. When the sensor and the model are both open, the durable layer is what OpenGait publishes here — the cited thresholds, the meaningful-change rule, and a named clinician accountable for the read.
Reference basis: Studenski et al., JAMA 2011; CDC STEADI; MOBILIZE Boston (U-shaped falls curve). Bands are educational reference points, not a diagnosis.
A conformant device emits this. A conformant consumer (an app, an EHR, the MotionSole engine, or your own) reads it. POST it to any OpenGait endpoint, or write it to a file.
{
"spec": "opengait/0.1",
"subject_ref": "anon-7f3a", // pseudonymous
"captured_at": "2026-06-29T14:22:00Z",
"capture_method": "imu_insole",
"gait": {
"gait_speed": 1.04, // m/s
"cadence": 102,
"stride_length": 1.22,
"symmetry": 91,
"double_support": 27.4,
"variability": 4.8,
"confidence": 0.93
},
"risk": { // optional, from §03
"band": "elevated",
"trend": "declining"
}
}Full JSON Schema, the FHIR Observation mapping, and the test vectors are provided to implementers — this page is the human-readable standard.
Being honest about the line is what keeps the spec trustworthy and the business viable. OpenGait is the language; the things people build with it can be open or commercial.
Implement it freely — build a product or a business on it. Attribution only.
The spec, the schema, the reference designs — all on this page. Implementer questions welcome.
Breaking changes bump the major; a changelog tracks every revision.
Intent: graduate to a neutral working group as adoption grows.
Want to implement, certify a device, or contribute a capture method? Get in touch → · See the live camera reference at /live and the engine that consumes it at motionsole.com.
Talk to us about a pilot →