Volo: AI-Enhanced Driver Wellness Monitoring

#Working Video Prototype of the Flutter app:

Demo-Video.mp4

Our "Empathic Co-pilot" for the Volkswagen i.mobilothon 5.0

Event: Volkswagen i.mobilothon 5.0 (Student Track)

---

1. 🎯 The Problem

Current driver monitoring systems are one-dimensional. They are reactive, only responding to drowsiness after it's a critical problem. Worse, they are completely blind to high-stress issues like road rage or driver anxiety. They also ignore critical environmental context, like elevated CO2 levels in the cabin, which can directly cause fatigue and impaired cognitive function.

2. 💡 Our Solution: Volo

Volo is the industry's first "empathic co-pilot." It is an AI-powered system designed to create a holistic understanding of the driver's state.

We believe in prevention over cure. Our system proves this by validating three distinct components:

1. The Volo AI Core (Python): A real-time computer vision server that uses a standard webcam to analyze facial landmarks (for drowsiness) and emotions (for agitation). It broadcasts alerts via WebSocket.
2. Hardware Prototypes (Arduino): We built and validated two sensor circuits:
   • MQ3 Alcohol Sensor for pre-drive compliance checks.
   • MQ135 Air Quality Sensor for monitoring CO2 levels.
3. The HMI (Flutter App): A cross-platform mobile app that serves as the Human-Machine Interface. It receives live alerts, displays the driver's real-time status, and sends system notifications.

3. ✨ Key Features

• Real-time Drowsiness Detection: Uses OpenCV and Dlib to calculate Eye Aspect Ratio (EAR) from the driver's facial landmarks.
• Stress & Agitation Detection: Uses a Facial Emotion Recognition (FER) model to detect high-stress states like anger.
• Pre-Drive Alcohol Lockout: Our hardware prototype proves the concept of using an MQ3 sensor to prevent ignition if alcohol is detected.
• Cabin Air Quality Monitoring: The MQ135 prototype validates monitoring CO2 levels to prevent "situational" drowsiness.
• Live HMI Dashboard: The Flutter app provides a clean, immediate status update ("SAFE", "DROWSY", "AGITATED") and sends push notifications for alerts.
• Zero-Latency & Privacy-Focused: The AI Core runs on-device (at the "edge"), ensuring interventions are instant and sensitive facial data never leaves the vehicle.

4. 🛠️ Technology Stack & Architecture

Our MVP combines a Python AI backend with a Flutter frontend. The proposed production architecture is designed for an on-device processor.

Component	Technology
AI Core (Python)	Python 3.9+, OpenCV, Dlib, FER, TensorFlow, Asyncio, WebSockets
HMI (Flutter App)	Flutter, Dart, web_socket_channel, flutter_local_notifications, fl_chart
Hardware Prototypes	Arduino Uno, MQ3 Alcohol Sensor, MQ135 Air Quality Sensor
Proposed Full-Stack	PostgreSQL (Database), Redis (Cache), AWS (Cloud Training/Analytics)

Architecture

The system is built around the Volo AI Core, which fuses data from the in-cabin camera, sensors, and the Vehicle CAN Bus. This on-device processing guarantees zero-latency interventions and user privacy.

5. 🚀 Getting Started

This project is in two parts: the Python AI server and the Flutter HMI app. You must run both.

Prerequisites

• Python 3.9+
• Flutter SDK
• An IDE (like VS Code)
• A physical Android/iOS device (for testing with a real IP) or an Emulator

Part 1: Run the Python AI Core (Server)

1. Navigate to the folder:

   cd volo_mvp/python_ai_core

2. Install dependencies:

   pip install -r requirements.txt

3. Download Dlib Model: You MUST download the facial landmark predictor model (shape_predictor_68_face_landmarks.dat) from this link.

   • Unzip it and place the .dat file inside the python_ai_core folder.

4. Run the server:

   python volo_ai_core.py

   The server will start, load the AI models, and turn on your webcam. You will see WebSocket server started at ws://0.0.0.0:8765.

Part 2: Run the Flutter App (HMI)

1. Find your Computer's IP:

   • Windows: Open Command Prompt and type ipconfig. Find your "IPv4 Address" (e.g., 192.168.1.10).
   • Mac/Linux: Open a terminal and type ifconfig | grep inet.
   • Note: Your computer and your phone MUST be on the same Wi-Fi network.

2. Update the IP in Flutter:

   • Open the file: volo_mvp/flutter_app/lib/services/websocket_service.dart.
   • Go to the connect() method (around line 75).
   • Change the wsUrl to use your computer's IP:

     // const wsUrl = 'ws://10.0.2.2:8765'; // Emulator IP
     const wsUrl = 'ws://YOUR_COMPUTER_IP_HERE:8765'; // e.g., 'ws://192.168.1.10:8765'

3. Run the app:

   cd volo_mvp/flutter_app
   flutter pub get
   flutter run

Part 3: Test the Demo!

Look at your webcam. Close your eyes for ~3 seconds. You should see the Python console print "STATE CHANGE: DROWSY" and your Flutter app's "LIVE STATUS" card will instantly turn orange and send a notification.

6. 🛣️ Future Scope

Our MVP validates the core AI and hardware concepts. The next steps are:

• Train the CNN-LSTM Model: Collect 200+ hours of fusion data (video, CAN bus signals) to train the full production model.
• Deep CAN Bus Integration: Move beyond prototypes to read live steering angle, pedal pressure, and vehicle speed from the CAN Bus[cite: 234].
• Automotive-Grade Sensors: Upgrade from Arduino to production-ready components, including Force-Sensing Resistors (FSRs) in the steering wheel for grip analysis.