v2.0 prototype
Ankle IMU modules
Dual inertial measurement units worn on the ankles to capture high-resolution kinematic data of the lower extremities.
- Dimensions
- 4.0 × 6.3 cm
- Sampling
- 100 Hz
- Placement
- Bilateral ankle
Technology & innovation
The instrument behind
the intervention
Built on research and clinical validation, FoGO Health pairs precise wearable sensing with an edge-AI pipeline light enough to run on the body.
Hardware specifications
Two coordinated modules: sensing at the ankle, computing and cueing at the chest.
v2.0 prototype
Chest vibrotactile module
The central processing unit and haptic feedback actuator, worn on the chest for optimal perceptual sensitivity.
- Perceptual threshold
- 0.72
- Function
- Compute + cue
- Placement
- Sternal, chest-worn
0.72
Why the chest?
Clinical testing found the chest offers superior perceptual sensitivity, a 0.72 threshold, compared with peripheral locations. The vibrotactile cue is reliably felt even during complex motor tasks or sensory overload, raising the success rate in overcoming a freeze.
Prototype worn during clinical testing
Intelligence that runs on the body
A pipeline light enough to classify gait state on a microcontroller, with no cloud round trip.
01
Feature extraction
Time and frequency domain features are extracted from raw IMU data in 2-second sliding windows.
02
Machine learning classification
A lightweight Random Forest model, optimized for microcontrollers, classifies the current gait state.
03
Ultra-low latency
The full processing pipeline executes in under 250 ms, allowing intervention before a fall occurs.
Latency budget~147 ms total
Preprocessing12 msButterworth filter
Feature extraction45 ms42 features
Inference85 msRandom Forest
Control logic5 msTrigger cue + log
edge_ai_pipeline.c
// Edge AI pipeline: executes on-device, per window
void process_imu_window(float* data) {
// 1. Preprocessing 12 ms
apply_butterworth_filter(data);
// 2. Feature extraction 45 ms
float features[42];
extract_freezing_index(data, &features[0]);
extract_energy_band(data, &features[1]);
// 3. Inference 85 ms
float fog_probability = rf_predict(features);
// 4. Control logic 5 ms
if (fog_probability > THRESHOLD) {
trigger_haptic_actuator(PATTERN_A);
log_event(TIMESTAMP, fog_probability);
}
}