SafeReps is a dual-stream coaching ecosystem that bridges the gap between following a workout video and having a personal trainer standing in the room. By fusing phone-based computer vision with a high-fidelity wearable sensor, SafeReps ensures every repetition is safe, effective, and counted with precision.

When you work out alone at home, you're "training blind." Workout videos can't see you, and static apps can't correct your form. SafeReps solves this by building a Digital Twin of your performance.

It detects the "invisible" physics of a rep—muscle tremors and momentum cheating—that no camera can catch alone. The moment your form degrades, the AI voice coach fires immediately to correct you mid-set.

• Dual-Stream Sensor Fusion: Merges 30 FPS vision landmarks with 100Hz high-fidelity IMU data.
• Invisible Fatigue Detection: Catch neuromuscular tremors before you feel them to prevent injury.
• Cheat Detection: Distinguishes between clean muscle contraction and momentum-based swinging.
• AI Voice Coach: Priority-gated audio feedback that provides corrections exactly when they happen.
• T-Pose Auto-Calibration: 1-second routine that aligns the wearable to your specific limb geometry.
• Hardware Economics: A professional coaching system built on a $5 BOM, making a $50 retail price possible.

SafeReps is designed for accessibility. Our prototype costs under $5 in components, proving that coaching-grade hardware doesn't have to be a luxury product.

• ESP32-C3: Logic & Bluetooth connectivity.
• MPU6050: 6-axis inertial measurement unit.
• 400mAh LiPo: Portable power for 12+ hours of active training.
• USB-C Module: Integrated charging.
• Protection Circuit: 100k voltage divider (monitoring) + Diode & Capacitor (safety).

graph LR

subgraph PWR ["Power & charging"]
  BAT["400mAh LiPo<br/><i>+ / −</i>"]
  CHG["USB-C charger<br/><i>B+ / G−</i>"]
  SW["Slide switch"]
  D["Diode<br/><i>A → K</i>"]
  CAP["100µF cap"]
end

subgraph VDIV ["Battery monitor"]
  R1["R1 · 100kΩ"]
  R2["R2 · 100kΩ"]
end

subgraph MCU ["Control"]
  ESP["ESP32-C3<br/><i>VIN · GND · GPIO 1</i>"]
end

subgraph SENSE ["Sensing"]
  MPU["MPU6050 IMU<br/><i>VCC · GND · SDA · SCL · INT</i>"]
end

BAT -- "B+" --> CHG
BAT -- "B−" --> CHG
BAT -- "+" --> SW
SW --> D
D -- "VIN" --> ESP
D -- "VIN" --> CAP
CAP -- "GND" --> BAT

BAT -- "+" --> R1
R1 --> R2
R2 -- "GND" --> BAT
R1 -- "GPIO 1 ADC" --> ESP

ESP -- "3.3V" --> MPU
ESP -- "SDA GPIO 8" --> MPU
ESP -- "SCL GPIO 9" --> MPU
ESP -- "INT GPIO 10" --> MPU

SafeReps manages a Finite State Machine (FSM) for every set to ensure movement is anatomically complete. Transitions are triggered by joint angles crossing calibrated thresholds, ensuring reps are only counted when they reach full range.

The ESP32-C3 wearable performs real-time Digital Signal Processing (DSP) before data hit the app:

• Tremor Analysis: A 100Hz high-pass filter isolates neuromuscular jitter from intentional movement.
• Cheat Detection: Calculates the ratio of Angular Velocity to Linear Acceleration to catch momentum-based swings.

Accuracy starts with alignment. SafeReps requires a 1-second T-Pose before every set. This enables:

• Sensor Zeroing: Synchronizes the wearable’s orientation to your skeletal model.
• Scaption Alignment: Defines the reference plane for your specific biomechanics, correcting for mounting tilt.

Follow the wiring diagram in the hardware section. Use the safereps-esp directory for the firmware source.

1. Navigate to safereps-esp/.
2. Use PlatformIO to upload: pio run -t upload.

1. Ensure the Flutter SDK is installed.
2. Navigate to safereps/.
3. Run flutter pub get followed by flutter run. (Note: Use a physical device for full BLE and Camera support.)

• LiDAR-Enhanced Tracking: Integrating front-facing LiDAR for sub-centimeter skeletal depth sensing.
• Shadow Boxing: High-speed strike velocity and "snap" analysis for combat sports.
• AR Overlays: Projecting "ghost reps" over the camera view in real-time.
• Physical Therapy: Expanding the library for home-based rehabilitation and recovery tracking.
• Full-Body Fusion: Multi-sensor support for complex compound movements (Squats/Deadlifts).

ble · c++ · computer-vision · dart · dsp · esp32 · flutter · ml-kit · mpu6050 · platformio

Built for MariHacks IX
View on Devpost