Kitchen Safety AI with Arduino UNO Q and OpenClaw

Arduino App Lab

Arduino® UNO™ Q 4GB

<html><body>In countries where gas stoves are common, a well-documented compulsion associated with OCD is repeatedly checking the knobs to confirm they are closed. People wake up in the middle of the night or drive back home just to verify that the burners are off — sometimes multiple times in a row. Is it possible to build an AI device to check the stove knobs? How complicated that is? Is it required to use an expensive board? <h3>The solution</h3><figure><img src="https://projects.arduinocontent.cc/967303ab-c680-4393-b2ec-96bdc472fdf3.png"/><figcaption>The Solution</figcaption></figure>The device uses two independent methods to assess stove safety: <ol><li>A USB camera aimed at the stove knobs, analyzed by the OpenClaw AI agent using its built-in vision model. </li><li>An MQ-2 gas sensor that cross-checks the visual result with an analog gas reading. </li></ol>The result is available on demand or on a schedule via Telegram or WhatsApp, in plain natural language. No app to open, no screen to look at. <figure><img src="https://projects.arduinocontent.cc/a7151b2c-f332-4f3f-91e6-0898403d5ab7.png"/><figcaption></figcaption></figure> Note: for environments where offline inference is preferred, Edge Impulse can be used to train and deploy a local model. <h3>Hardware</h3><h3></h3><figure><img src="https://projects.arduinocontent.cc/f8f3ed5e-24da-4b38-ba23-9135d1b431f4.jpg"/><figcaption>UNO Q</figcaption></figure> The core of this project is the Arduino UNO Q — a significant departure from previous UNO boards. It has a dual-processor architecture: STM32U585 MCU: Runs Arduino sketches on Zephyr OS — handles sensor reading and real-time tasks Qualcomm Dragonwing QRB2210 MPU: Runs Debian Linux — hosts OpenClaw, Python scripts, and networking This means the board can simultaneously run complex software like an AI agent and perform real-time hardware tasks like reading a sensor — something previous UNO boards were not designed for. The UNO Q comes in two versions: 2 GB and 4 GB RAM. This project uses the 4 GB version. <h3>Full Parts List</h3> <figure><img src="https://projects.arduinocontent.cc/afb76361-410b-46f4-bd3c-a300d0f2c42e.jpg"/><figcaption></figcaption></figure> <ol><li><a href="https://amzn.to/42oNcJh ">Arduino UNO Q (4 GB recommended) </a></li><li><a href="https://amzn.to/3Ph2Ymj ">Logitech USB webcam (or any other) </a></li><li><a href="https://amzn.to/3Pzghib ">USB hub with power delivery (the UNO Q has a single USB-C port; you need it for power and the camera simultaneously) </a></li><li><a href="https://amzn.to/4dhBTYl ">MQ-2 gas sensor module (sometimes sold as FC-22) </a></li><li>3 x Jumper wires female-male </li></ol><h3>Circuit</h3><h3></h3><figure><img src="https://projects.arduinocontent.cc/e4884e56-e753-44d0-a78c-8e1f0d3ff74c.jpg"/><figcaption>Cables</figcaption></figure> The wiring is minimal. <ol><li>UNO A0 to MQ2 Aout </li><li>UNO 5V to MQ2 VCC </li><li>UNO GND to MQ2 GND </li></ol>The ADC resolution on the UNO Q can be configured between 8, 10, 12, and 14 bits using analogReadResolution(bits) in the sketch. The default used in this project is 10 bits (values 0–1023). There is no universal threshold value for the MQ-2. The analog reading in clean air varies between individual sensor modules, ambient temperature, humidity, and altitude. A hardcoded number is not reliable. The correct approach is to measure a baseline in clean air and alert on a significant rise above it: <ol><li>Power on the device and wait at least 2 minutes. MQ-type sensors require warmup time for the internal heater element to reach operating temperature. Readings taken immediately after boot are not meaningful. </li><li>In a confirmed gas-free environment, take readings every second for 60 seconds and average them. This is the clean-air baseline. </li><li>Set the alert threshold to approximately 130% of the baseline (a 30% rise). Tighten to 120% in sensitive environments or loosen to 150% if cooking smells or aerosols cause false positives. </li></ol><h3>3d Enclosure</h3> <figure><img src="https://projects.arduinocontent.cc/7ffd173e-9c13-483f-b7a9-8e8b7d7e5c91.png"/><figcaption>Enclosure</figcaption></figure>A basic enclosure is provided with this project so the device can be mounted on any standard camera tripod and positioned in front of the stove knobs. Print in PLA with supports enabled. <h3>Image recognition</h3>Depending on the use case, image recognition may require training an ML model. One advantage is that it is possible to use images from our specific kitchen under different lighting conditions. <a href="https://www.edgeimpulse.com/">Edge Impulse can be used, and the model can be deployed locally, even without an Internet connection </a>. The downside is the training and deployment process. In this case, the decision was to use OpenClaw with openai-codex/gpt-5.5 model. <figure><img src="https://projects.arduinocontent.cc/f7970738-08b3-4370-aaf9-ee5580fae13f.png"/><figcaption>OpenClaw</figcaption></figure><h3> Gas sensor readings</h3>The STM32U585 MCU and the Dragonwing QRB2210 MPU are separate processors. They communicate through the Arduino RouterBridge — an RPC (Remote Procedure Call) mechanism built into the UNO Q. On the MCU side (sketch.ino), a function is registered: <pre>Bridge.begin(); </pre><pre>Bridge.provide("readSensor", readSensor); </pre>On the Linux side (check.py), Python calls it by name: <pre>from arduino.app_utils import Bridge </pre><pre>value = Bridge.call("readSensor") </pre>The Bridge handles serialization and transport between the two processors transparently. The Arduino UNO Q Led Matrix is used to display the gas readings. Each column will display <h3>Software setup</h3>1. Configure Wi-Fi and SSH Debian Linux comes pre-installed on the UNO Q. To configure Wi-Fi and enable SSH access, install the Arduino App Lab application on your computer, connect the board, and follow the setup wizard to enter your Wi-Fi credentials and set a password. <figure><img src="https://projects.arduinocontent.cc/c2fad3fc-9555-451f-8dfd-1d8352aa7c51.png"/><figcaption></figcaption></figure>Once configured, connect via SSH using any SSH client (such as PuTTY on Windows): <pre>Host: &lt;UNO Q IP address&gt; or arduino@BOARDNAME </pre><pre>User: arduino </pre><pre>Password: &lt;the password you set in App Lab&gt; </pre> 2. Install OpenClaw <figure><img src="https://projects.arduinocontent.cc/6aa55d26-6ceb-4ad3-9b1d-b70634bf0007.png"/><figcaption></figcaption></figure>OpenClaw is the AI agent that runs on the Linux side of the UNO Q. Install it with: <pre>curl -fsSL https://openclaw.ai/install.sh | bash </pre>OpenClaw supports multiple AI model providers and messaging channels. This project uses OpenAI and Telegram, but the choice is yours. <h3>Get Telegram Token and Channel</h3><ol><li>Open Telegram and search for @BotFather. </li><li>Send /newbot and follow the prompts to create a bot. Save the token you receive. </li><li>Search for your new bot and click Start. </li><li>Send a message to @userinfobot to get your personal chat ID. </li><li>Enter the token and chat ID into the OpenClaw configuration. </li></ol><h3>Configure the Webcam</h3> <pre>sudo apt-get update -y </pre><pre>sudo apt-get install -y fswebcam v4l-utils </pre><pre>sudo usermod -aG video arduino </pre>Adjust camera settings for better image quality in typical kitchen lighting: <pre>v4l2-ctl -d /dev/video0 \ </pre><pre>	 --set-ctrl=brightness=180 \ </pre><pre>	 --set-ctrl=contrast=150 \ </pre><pre>	 --set-ctrl=saturation=140 \ </pre><pre>	 --set-ctrl=gain=200 \ </pre><pre>	 --set-ctrl=backlight_compensation=1 </pre>Take a test photo to confirm the camera is working and aimed correctly: <pre>fswebcam -d /dev/video0 -r 1280x720 --no-banner test.jpg </pre>Retrieve the image over SFTP (using FileZilla or similar) to inspect it on your computer. Download <a href="https://github.com/ronibandini/UNOQOpenClawAntiOCD/blob/main/antiocd.zip">https://github.com/ronibandini/UNOQOpenClawAntiOCD/blob/main/antiocd.zip</a>Upload the sketch.ino and the .py using AppLab Select Run at startup <figure><img src="https://projects.arduinocontent.cc/9af94b0d-0c5c-407a-b8fa-3a111cbf4e7d.png"/><figcaption>Applab</figcaption></figure>Upload the OpenClaw skill to /home/arduino/.npm-global/lib/node_modules/openclaw/dist/extensions/browser/skills/ocd-knob-routine <h3>Using the Device</h3><figure><img src="https://projects.arduinocontent.cc/62457ff6-ae87-4744-b27c-5fd38483e54f.png"/><figcaption>Telegram</figcaption></figure>Once everything is set up, just send a message to your Telegram bot in plain language: • “Is the stove off?” • “Did I leave the stove on?” • “Check the gas.” • “Check the knobs before I go to sleep.” OpenClaw will run the check, analyze the image with its model, analyze gas sensor readings, and reply with a clear status message. You can also ask it to schedule a check automatically at a set time each night. <figure><img src="https://projects.arduinocontent.cc/c76f0b79-2e48-4eef-ba26-7d72446a3306.png"/><figcaption>Prototype</figcaption></figure><h3>Final notes</h3>This project was, before anything else, an excuse to try the new Arduino UNO Q. It is surprisingly easy to get running — App Lab handles Wi-Fi setup and that’s essentially it. The board is capable enough to run complex software like an AI agent while remaining fully compatible with the classic Arduino ecosystem of libraries and components. Possible enhancements: <ol><li>Persistent logs and reports. </li><li>An LED ring or lamp to illuminate the stove area for better nighttime image quality. </li><li>A local ML model via Edge Impulse for fully offline knob detection (and to save LLM tokens) </li></ol><h3>Video</h3><figure><iframe src="https://www.youtube.com/watch?v=xSUeZptGj98"></iframe></figure><h3>References</h3>https://www.arduino.cc/product-uno-q <a href="https://docs.arduino.cc/tutorials/uno-q/routerbridge-multilanguage/">https://docs.arduino.cc/tutorials/uno-q/routerbridge-multilanguage/ </a>https://docs.arduino.cc/software/app-lab/getting-started/quickstart/ <h3>Contact</h3>Roni Bandini <a href="https://www.instagram.com/ronibandini/">https://www.instagram.com/ronibandini/</a><a href="https://x.com/RoniBandini">https://x.com/RoniBandini</a><a href="https://www.linkedin.com/in/ronibandini/">https://www.linkedin.com/in/ronibandini/</a></body></html>

Kitchen Safety AI with Arduino UNO Q and OpenClaw

An AI-powered device that verifies stove safety using computer vision, gas sensing, and natural language through Telegram