Inspiration
Freezing of gait is a Parkinson's symptom where the feet stick to the floor mid-step while the body keeps moving. It affects up to 70% of patients after a decade with the disease and is a leading cause of falls.
Cueing already works. A Neurology study found laser-projecting shoes cut freezing by 38 to 46%. But they fire on every step. Nobody closed the loop: detect the freeze first, then cue.
What it does
An ankle wearable samples motion at 64 Hz and detects a freeze on-device: freezing shows up as 3 to 8 Hz energy that normal walking does not have.
On detection it fires cues instantly, with no network in the loop: vibration on the shoe, a beat at walking cadence, and a spoken prompt. A bandit learns which cue works fastest for that person.
Every episode is logged with its duration and the cue that broke it.
How we built it
- ESP32-S3 and MPU-6050 on an ankle strap, 64 Hz
- FFT and freeze index computed on-chip
- Vibration motor, buzzer and LEDs, and a spoken prompt
- Thompson sampling bandit, rewarded by time to resume walking
- Dashboard with the episode log
Validation
We tested the detector against Daphnet, a public dataset of 10 Parkinson's patients with 237 expert-labeled freezes recorded at 64 Hz. That let us set thresholds against real patient data instead of guessing.
Challenges we ran into
The motor shakes the same strap as the sensor, so its buzz looked like a freeze. We moved the motor to the shoe and filtered at 20 Hz.
Standing still is not freezing. Our first version flagged anyone who stopped walking, so we added a movement requirement.
Log in or sign up for Devpost to join the conversation.