Tutorial 17 — MQTT Integration: Connect Jetson AI to Factory Systems

Connect your Jetson AI vision system to the rest of your factory. Use MQTT to send real-time defect alerts, occupancy counts, and machine events to PLCs, SCADA systems, dashboards, and mobile apps — with sub-100ms latency.

What you will learn

• What MQTT is and why it is the standard for IoT/factory messaging
• How to publish AI detection events to an MQTT broker
• How to subscribe to topics from a dashboard, PLC, or app
• How to structure MQTT payloads for maximum compatibility
• How to set up a local Mosquitto broker (pre-installed on your kit)

Step 1 — Start the Mosquitto broker

# Mosquitto MQTT broker is pre-installed
sudo systemctl start mosquitto
sudo systemctl enable mosquitto   # auto-start on boot

# Test it works
mosquitto_sub -t "test/#" -v &
mosquitto_pub -t "test/hello" -m "Jetson connected!"

Step 2 — Publish AI alerts from Python

import paho.mqtt.client as mqtt
import json, time

client = mqtt.Client("jetson-vision")
client.connect("localhost", 1883)
client.loop_start()

def publish_detection(camera_id, class_name, confidence, bbox):
    payload = json.dumps({
        "timestamp":  time.time(),
        "camera_id":  camera_id,
        "class":      class_name,
        "confidence": round(confidence, 3),
        "bbox":       bbox,
        "alert":      confidence > 0.85
    })
    client.publish(f"vision/{camera_id}/detection", payload)

# Call this every time a defect is detected:
publish_detection("cam1", "scratch", 0.92, [120, 80, 340, 210])

Step 3 — Subscribe from Node-RED dashboard

# Node-RED is pre-installed at http://localhost:1880
# Add an MQTT-in node, set broker to localhost:1883
# Topic: vision/+/detection
# Connect to a dashboard gauge, alert, or database node

Step 4 — Run the full MQTT demo

cd ~/tutorials/17-mqtt-integration
python3 mqtt_vision.py --source 0 --broker localhost --show

✅ Next: Tutorial 18 — Grafana Dashboard | Back to Jetson Kit