Hybrid, Ultra-silent, Self-Healing Drone Mesh

Overview

HUSH-MESH is an AWS-first maritime threat detection and convoy protection system leveraging machine learning, edge computing, and real-time path planning. The system deploys autonomous drone networks for defensive surveillance and threat assessment in maritime environments.

Key Capabilities: Real-time threat detection • Multi-sensor fusion • Edge ML inference • Self-healing mesh networks • Dynamic route optimization

---

Quick Start

Prerequisites

Ensure you have Python 3.9+ installed:

python3 --version

Install required dependencies:

pip install flask flask-cors fastapi uvicorn paho-mqtt torch numpy

Starting the System

1. Reset Any Existing Processes

pkill -f "python.*server" && pkill -f "python.*app" && pkill -f "http.server"

2. Start ML API Server (Port 9000)

cd ml
export PYTHONPATH=/home/participant/.local/lib/python3.11/site-packages:$PYTHONPATH
python3 ml_api_server.py > /tmp/ml_server.log 2>&1 &

3. Start Demo Server (Port 8081)

cd demo
python3 -m http.server 8081 > /tmp/demo_server.log 2>&1 &

Verify System Status

Check that all servers are running:

# Check ML API Server
curl http://localhost:9000/health

# Check Demo Server
curl http://localhost:8081/

# View server logs
tail -f /tmp/ml_server.log
tail -f /tmp/demo_server.log

---

Access Points

Interactive Demos

• Enhanced Multi-Route Demo: http://localhost:8081/enhanced_multi_route.html
• ML Test Interface: http://localhost:8081/test_ml.html

API Endpoints

ML API Server (Port 9000)

• GET /health – Health check and server status
• POST /predict – Threat prediction inference
• POST /reset – Reset model state

Demo Server (Port 8081)

• Serves static HTML demos and visualizations

---

System Architecture

┌─────────────────┐    ┌─────────────────┐    ┌─────────────────┐
│   Edge Layer    │    │   Cloud Layer   │    │  Dashboard UI   │
│                 │    │                 │    │                 │
│ • Greengrass    │◄──►│ • IoT Core      │◄──►│ • React App     │
│ • ML Models     │    │ • Kinesis       │    │ • Real-time Map │
│ • Sensors       │    │ • Lambda        │    │ • WebSocket     │
│ • MQTT Comms    │    │ • DynamoDB      │    │ • Cognito       │
└─────────────────┘    └─────────────────┘    └─────────────────┘

Core Technologies:

• Edge Computing: AWS IoT Greengrass + SageMaker Neo optimized models
• Cloud Infrastructure: Kinesis Streams + Lambda + DynamoDB + ECS Fargate
• ML Pipeline: SageMaker training with maritime threat detection models
• Communication: MQTT mesh networking for resilient edge coordination

---

Development Workflow

Running Tests

# Test ML inference
python3 test_ml_integration.py

# Test convoy simulation
python3 simple_demo.py

Monitoring Logs

# Monitor ML server
tail -f /tmp/ml_server.log

# Monitor demo server
tail -f /tmp/demo_server.log

# Check for errors
grep -i error /tmp/ml_server.log

Troubleshooting

Servers Won't Start

# Check port availability
netstat -tlnp | grep -E ":(8081|9000)"

# Force kill and restart
pkill -9 -f "python.*server"
pkill -9 -f "http.server"

# Restart servers (run commands from steps 2-3 above)

ML Model Issues

# Verify model file exists
ls -lh models/model.pth

# Check Python path
echo $PYTHONPATH

# Reinstall dependencies
pip install --force-reinstall torch numpy

Connection Refused Errors

Ensure all servers are running and listening on correct ports:

ps aux | grep python
netstat -tlnp | grep LISTEN

---

Deployment to AWS

For production deployment with full AWS integration:

1. Configure AWS CLI with appropriate credentials
2. Deploy infrastructure using CDK/CloudFormation
3. Set up IoT Greengrass on edge devices
4. Deploy ML models to SageMaker endpoints
5. Configure Kinesis streams for telemetry ingestion

See docs/deploy_instructions.md for comprehensive deployment guide.

---

Project Structure

hush-mesh/
├── ml/                     # ML inference server
│   ├── ml_api_server.py   # Flask API for threat detection
│   └── models/            # Trained PyTorch models
├── demo/                   # HTML demos and visualizations
│   ├── enhanced_multi_route.html
│   └── test_ml.html
├── backend/               # Backend API server (optional)
├── docs/                  # Documentation
│   ├── architecture.md
│   └── deploy_instructions.md
└── README.md

---

Security & Ethics

Human-in-the-Loop Design: All engagement decisions require explicit human authorization. The system provides threat assessment and recommendations only.

Defensive Posture: System is designed exclusively for defensive maritime convoy protection. No offensive capabilities.

Data Privacy: Synthetic training data only. No PII collection. All telemetry encrypted in transit and at rest.

---

License

MIT License

---

Developed for: AWS Mission Autonomy Hackathon Fall 2025 Team: Clankers Status: Active Development – MVP Complete