⚡ HabiZap

⚠️ Disclaimer / Important Notice

This repository contains a personal hobby project and early-stage development prototype.

• This is NOT a medical device.
• This project is NOT safety-certified.
• It is NOT intended for diagnosis, treatment, mitigation, or prevention of any medical, psychological, or health condition.
• You use, modify, build, or distribute this project entirely at your own risk.

The author provides this project "AS IS", without warranty of any kind, express or implied, including but not limited to fitness for a particular purpose. In no event shall the author be held liable for any claim, damage, injury, loss of data, or legal consequence arising from the use of this project.

This repository exists strictly for learning, experimentation, and research purposes.

---

Overview

HabiZap is an experimental wearable wristband project focused on habit interruption using on-device machine learning.

Instead of passively tracking metrics or showing notifications, HabiZap attempts to:

• Detect repetitive or harmful hand gestures (e.g. nail biting, hair pulling)
• Classify them locally using ML inference
• Respond immediately with haptic feedback to interrupt the habit loop

The project intentionally avoids displays and cloud dependencies to remain:

• Minimal
• Private
• Power-efficient
• Hackable

---

Credits

• 3D prototype design: Special thanks to MaryBR for designing the early 3D enclosure prototype.

  • LinkedIn: https://www.linkedin.com/in/mary-br/

The STL file will be shared soon in this repository.

---

Roadmap

• Prototype of with the XIAO ESP32S3 + MPU6050 + vibration motor
• Basic ML integration with Edge Impulse
• Basic gesture detection and haptic feedback
• Battery voltage monitoring
• Design a custom PCB and small case for it
• Custom App for the wristband
• Auto train the ML without manual Edge-Impulse training

---

Hardware at a Glance

ESP32-S3 MCU, MPU6050 IMU, vibration motor module, and single-cell Li‑Po battery

---

Philosophy & Design Principles

• Wristband-first: small, lightweight, unobtrusive
• No display: reduces distraction and complexity
• Local-first ML: inference happens on-device
• No mandatory internet: no always-on cloud
• Prototype-friendly: optimized for iteration, not polish

---

1. Specifications

1.1 Hardware

1.1.1 MCU – XIAO ESP32S3

The project uses XIAO ESP32S3 as the main MCU due to:

• Integrated single-cell Li‑Po battery charger
• Sufficient compute for ML inference
• Vector instructions useful for ML workloads
• Very small physical footprint

Reference: https://wiki.seeedstudio.com/xiao_esp32s3_getting_started/

---

1.1.2 Motion Sensor – MPU6050 (GY‑521)

The prototype uses an MPU‑6050 module for motion data:

• 3‑axis accelerometer
• 3‑axis gyroscope
• I²C interface
• 1024‑byte FIFO buffer

Connections:

• SDA → ESP32 SDA
• SCL → ESP32 SCL
• VCC → 3.3V
• GND → GND
• AD0 → GND (I²C address 0x68)

Unused pins:

• INT, XDA, XCL

---

1.1.3 Vibration Motor Module

A vibration motor module provides haptic feedback when a target gesture is detected.

Typical specs:

• Operating voltage: 3.0 – 5.3 V
• Rated current: ~60 mA
• Startup current: ~90 mA

⚠️ The vibration motor is powered directly from the battery, not from the ESP32 regulator.

The module already includes a MOSFET and required driver components.

Pins:

• VCC → Battery +
• IN → ESP32 PWM pin
• GND → Ground

---

1.1.4 Battery & Power Monitoring

• Single‑cell Li‑Po battery
• Capacity: ~150–700 mAh

Charging is handled by the XIAO ESP32S3 internal charger.

Because there is no integrated fuel gauge, battery voltage is measured using a simple ADC voltage divider:

• R1: 47kΩ
• R2: 100kΩ

Reference: https://wiki.seeedstudio.com/check_battery_voltage/

---

1.1.5 Bill of Materials (Prototype)

ID	Component	Value	Purpose
R1	Resistor	47kΩ	Battery voltage divider
R2	Resistor	100kΩ	Battery voltage divider
–	MPU6050	GY-521	Motion sensing
–	Vibration motor	Module	Haptic feedback
–	Li-Po battery	150–700 mAh	Power source

---

1.1.6 Schematic

Prototype-level schematic (power, IMU, vibration motor, and ADC voltage divider)

---

1.2 Firmware

• Framework: ESP‑IDF
• OS: FreeRTOS
• Language: C (core firmware)
• ML inference: C/C++ code generated by Edge Impulse

Firmware responsibilities:

• Sensor sampling and buffering
• Task scheduling via FreeRTOS
• ML inference execution
• Gesture classification
• Haptic feedback control
• Optional streaming for training

---

1.3 Kconfig Overview

The project uses Kconfig (via ESP‑IDF) to configure build‑time behavior.

Key configuration options include:

• Training Mode Enable

  • Enables raw sensor data streaming for ML data collection
  • Disables inference‑only optimizations

• Sampling Configuration

  • Motion sensor sample rate
  • Buffer sizes

• Debug / Logging Options

  • Serial logging level
  • Development diagnostics

You can access the configuration menu using:

idf.py menuconfig

⚠️ Make sure to disable training mode for normal inference usage to reduce power consumption.

---

1.4 Mechanics

• Prototype stage
• Off‑the‑shelf modules
• No custom enclosure yet

Future revisions will introduce:

• Custom PCB
• Smaller footprint
• Wristband‑friendly enclosure

---

1.5 Companion App

🚧 Not implemented yet

Planned responsibilities:

• Device configuration
• Training session control
• Gesture labeling
• Sensitivity tuning
• OTA firmware updates

---

2. Machine Learning Workflow

2.1 Training with Edge Impulse

The current ML workflow relies on Edge Impulse for data collection and model training.

Prerequisites

• Edge Impulse account
• Node.js
• ESP‑IDF toolchain installed

---

Training Steps

1. Create a new Edge Impulse project
2. Install the Edge Impulse CLI:

npm install -g edge-impulse-cli

3. Enable training mode in firmware:

idf.py menuconfig

Enable the HabiZap training option.

4. Build and flash the firmware
5. Close the serial monitor
6. Run the data forwarder:

edge-impulse-data-forwarder

7. When prompted for sensor labels, enter:

ax,ay,az,gx,gy,gz

8. Collect data via the Edge Impulse dashboard

---

2.2 Integrating the Trained Model

After training and deploying your model from Edge Impulse:

1. Download the C/C++ library generated by Edge Impulse
2. Extract and copy the generated files into the root of this project
3. Ensure the Edge Impulse source files are included in the build
4. Rebuild the firmware

⚠️ This step is required. The project does not fetch Edge Impulse artifacts automatically.

---

Project Status

🚧 Active experimental prototype

Expect:

• Breaking changes
• Incomplete features
• Hardware revisions

---

License

This project is released for personal, educational, and experimental use.

Refer to the LICENSE file for details.

---

Final Notes

HabiZap is an exploration of:

• Embedded ML
• Wearable hardware design
• Gesture recognition
• Minimalist product thinking

If you build upon this project, please do so responsibly and ethically.