I've been tinkering with single-board computers and microcontrollers for a while, and one thing took me way too long to realize: they're not competitors — they're better together. An Arduino is great for circuits, sensors, and automation, while a Raspberry Pi handles the heavier computing. Combine them and you unlock projects neither can do well alone.
My favorite combo is a Home Assistant weather station. The Arduino reads a BME280 sensor on a breadboard, while the Raspberry Pi runs Home Assistant, and I link the two over MQTT. It's cheap, reliable, and the dashboard shows live temperature and humidity from my balcony.
Another one I keep coming back to is a smart chess board: an Arduino Nano drives the LEDs and switches under the board, while the Pi runs the Stockfish chess engine for an AI opponent. It's a fun mix of electronics, 3D printing, and software.
If you're into retro gaming, a Raspberry Pi makes a brilliant emulation box, and an Arduino can become a DIY game controller for it. I've also seen clever folks build an RFID door lock where the Pi stores the tag database and the Arduino handles the reader and strike — a genuinely useful home project.
If you're just starting, don't overthink it: grab a used Arduino and a Pi, and try one small project end to end. The lessons compound fast, and before you know it you'll be dreaming up your own builds.
Source: XDA Developers
This project, published on the Robotistan Maker blog, tackles a problem anyone who goes from meeting to meeting will recognize: instead of constantly checking the calendar tab or waiting for a phone notification, you see the next meeting on a physical screen sitting on your desk.
The project uses a 5.0-inch ESP32 HMI TFT LCD. The screen already includes an ESP32-S3 microcontroller, Wi-Fi and the graphics driver circuitry, so there is no need to wire a separate ESP32 with jumper cables. The system connects to Wi-Fi, queries the next event through the Google Calendar API, and shows the meeting title, time and description on the 800x480 display. The current time comes from an NTP server. When you add a new meeting to the calendar you don't need to touch the ESP32, because the device checks the calendar at regular intervals and updates the screen itself.
The screen is capacitive touch, but this version doesn't use touch at all; the panel is purely an information display. The project goes beyond electronics and software: it runs on a 3.7 V 1800 mAh 1S Li-Po battery, and a 3D-printed wall case holds both the screen and the battery. That lets the panel be mounted on a wall without staying plugged into a USB cable. During development a 30-pin ESP32 development board was also used, but it isn't required for the final build.
What I like is that it builds a genuinely useful everyday tool from very few parts. The same idea could be adapted for showing a school or workshop schedule. The parts list is short, so beginners can give it a try without much hassle.
Source: Robotistan Maker
Continuing its experiments with the RTL-SDR V4 dongle, framboise314 has published a guide to decoding APRS on a Raspberry Pi 4. The idea is to turn the small board into an amateur radio receiving station, listening on the European VHF frequency of 144.800 MHz.
The article starts by explaining what APRS (Automatic Packet Reporting System) is. Invented in the late 1980s, it is packet radio: very short bursts of audio in 1200-baud AFSK, wrapped in AX.25 frames. They are used for position tracking, telemetry (weather stations, the battery status of an isolated relay) and short messages. The network works without the Internet thanks to digipeaters, and iGates bridge it to the web. The author also compares it with Meshtastic, whose main difference lies in regulations and frequencies.
On the practical side, the setup goes through GQRX, tuned carefully, and then through Dire Wolf, a purely software TNC that decodes the frames. The tutorial covers the installation, routing audio over UDP, and a very real pitfall: GQRX sends over IPv6 by default, so the stream has to be forced to IPv4 for Dire Wolf to receive anything. Finally, it deals with a hardware problem, thermal throttling of the processor under the load of signal processing, which must be addressed if you want a station that runs reliably around the clock.
What I found pleasant is the tone of someone who practised packet radio in the 1990s and is rediscovering the technique with hardware that costs a few tens of euros. A good starting point for anyone who wants to combine Raspberry Pi and radio.
Source: Framboise 314
When people talk about LLMs they immediately think of chatbots, but Needle takes a different path. It is a small AI model, just 14 MB, developed by Cactus Compute, with one job: turning a natural-language request into a call to a Python function. It is not built for conversation. It is built to control applications, devices and automations.
What caught my attention is that it runs locally on a Raspberry Pi 5 using only the CPU, with no cloud and no dedicated AI hardware. Instead of writing accendi_led(), you just say "Turn on LED 1" and the model picks the right function. It also works with Italian commands, and the project's website has online demos (home automation, robot vacuum, smartwatch, AR glasses) you can try.
The RaspberryItaly tutorial builds a small local assistant that saves notes to a file, reads the Raspberry Pi's temperature, switches an LED on and off, and understands commands written in Italian. You need a Raspberry Pi (preferably a 5), an up-to-date Raspberry Pi OS, Python 3, an LED with a 220 or 330 ohm resistor and a few jumper wires. The installation goes through a virtual environment and the cactus-needle package; the internet connection is only needed for the first download.
One tip from the author is worth remembering: it is better to give one command at a time. A sentence like "turn off the kitchen light and lower the blind" may not be interpreted correctly, while the two commands given separately work.
As a project idea I find it very practical: combined with local speech recognition, it would let you command a robot or a small automation without depending on any network service.
Source: RaspberryItaly
Embarcados introduces the XIAOML Kit from Seeed Studio, made to follow the hands-on labs of the book «Machine Learning Systems» by Professor Vijay Janapa Reddi of Harvard. The labs were developed together with Professor Marcelo Rovai of UNIFEI in Itajubá, within the tinyML4D group.
The kit brings a XIAO ESP32-S3 Sense with an OV3660 camera, a digital microphone and an SD card reader, plus an expansion board with a 6-axis IMU and a 0.42-inch OLED display. That covers the three kinds of data the book works with: images, sound and motion. It also includes a 2.4 GHz FPC antenna, two heat sinks and a 32 GB microSD card with adapter.
The SoC is the ESP32-S3R8, a dual-core Xtensa LX7 at 240 MHz with 512 KB of SRAM, 8 MB of PSRAM and 8 MB of flash. The IMU is the LSM6DS3TR-C, and the OLED uses an SSD1315 controller, which the libraries treat as an SSD1306 with 72 x 40 pixels. The assembled unit measures 21 x 17.8 x 30 mm and weighs 10.6 g. A practical detail the author points out is that the IMU and the OLED share the same I2C bus at addresses 0x6A and 0x3C, so anyone adding another I2C sensor has to avoid those two. The IMU board also has a JST connector for a 3.7 V lithium battery, which is handy for wearable projects.
What I liked most is the mix of a reference book with cheap, compact hardware: you can move from theory to running inference on a real microcontroller. The OLED only fits two or three readable lines, of course, but that is enough to show the inferred class or a confidence percentage.
Source: Embarcados
Arduino is taking pre-orders for the Ventuno Q, which was already announced in March. According to a Make report, the board costs 298.99 euros, and pre-ordered units are expected to ship in about four weeks. It is aimed mainly at edge AI and robotics.
The "dual brain" concept is what I find exciting. On the Linux side there is a Qualcomm Dragonwing IQ8 (QCS8275) with eight Kryo cores and a Hexagon NPU delivering up to 40 TOPS. For everything time-critical there is an STM32H5F5 with a Cortex-M33, running Zephyr RTOS with the Arduino core and handling motors, PWM and GPIOs. The two sides talk to each other over an RPC bridge. A robot could analyze camera images locally and handle motor control at the same time, without combining several boards.
The spec is generous by Arduino standards: 16 GB LPDDR5, 64 GB eMMC, an M.2 slot for NVMe SSDs, three MIPI-CSI camera connectors, 2.5-gigabit Ethernet, Wi-Fi 6 and Bluetooth 5.3. Add I2C, SPI, UART, several CAN-FD interfaces, a Qwiic connector, four RGB LEDs and an 8x13 LED matrix.
On the software side, Ubuntu is planned and Debian will follow. According to Arduino, Docker and apt are ready to use. In the Arduino App Lab, sketches, Python scripts and containerized AI models can be connected. Preinstalled are Qwen, Gemma, Whisper, YOLOX and MediaPipe among others, and your own GGUF models run via llama.cpp.
The price is far above the usual Arduino level, so this board is better suited to larger projects than to beginners. What really appeals to me is the idea of local voice control and automation that doesn't depend on the cloud.
Source: heise online (Make)
During the summer heat, more and more elderly people living alone are being taken to hospital with heatstroke because they didn't notice the room getting hot. «THSense», a personal project published on Zenn, is aimed at people caring for family members who live far away. It keeps an eye on the room's temperature and humidity and sends an email when the temperature reaches a dangerous level.
The features are simple. It measures the room's temperature and humidity every minute, sends an email when it goes above 35 °C, and lets you see the current values and the history as a graph in the browser. According to the author, the graph is there so you can spot early signs, such as a temperature that keeps climbing.
The architecture is interesting. Reading the sensor (DHT20) is left to an Arduino on the Grove Beginner Kit rather than the Raspberry Pi, because it's better for real-time response and makes accurate readings easier. Every 60 seconds the Arduino sends a string over USB serial to the Raspberry Pi, where Python receives it with pyserial, turns it back into numbers, and handles the temperature check, sending the email, and saving to the database. The database is MariaDB on a virtual machine in Azure, and the screen is built with Flask and Chart.js. Emails go out through Gmail's SMTP server using smtplib, and the account details are read from a .env file instead of being written in the code.
The division of roles is clear, which makes it a useful reference for small IoT design. Sensing on the Arduino and judgment and notification on the Raspberry Pi is a split I could also use when thinking about similar builds. The author says the source code is published on GitHub.
Source: Zenn
Raspberry Pi has finally put its Smart Display Module on sale. It was first announced back in January, and as of this week you can buy it from Approved Resellers for $30. Compute Module 5 is sold separately, so it is a bare adapter rather than a whole computer.
What it is: an adapter board for Compute Module 5 that slots into displays and projectors following Intel's Smart Display Module (SDM) specification. Power, video, audio, data and control all travel through the internal connector, which means no extra cable, power supply or bracket hanging off the back of a screen. It was developed together with Sharp Display Solutions Europe and handles 4K video playback.
There is also an on-board M.2 slot, which can take fast storage or an AI accelerator such as a Hailo module. The board is fanless and silent, aimed at signage that has to run 24/7: flight information displays, wayfinding, retail and corporate screens, or immersive exhibits when paired with an SDM projector.
One early tester, a small-business signage service called Media Looper, ran it on a Sharp display for three months. Its founder said it does everything their Raspberry Pi 5 setup does, without the cable, power supply or bracket a customer could knock loose, and that they would make it the default for any display with an SDM slot.
I like this kind of announcement because it shows how far the platform has moved from hobby boards toward tidy, embedded hardware. For anyone building a display project, a $30 module that disappears inside the screen is a neat option, though you still need to check that your display actually has an SDM slot.
Source: Raspberry Pi News
The HDA Link interface, also called Azalia, is usually found only in PC chipsets. The author of a Habr article decided to check whether the RP2040 microcontroller on the Raspberry Pi Pico could act as its host. The appeal is clear: multichannel HDA audio codecs offer far more channels than I²S.
The project is built around an RP2040 with PIO blocks and a Realtek ALC889 codec on the board, a board left over from an earlier failed attempt with an STM32F103 and a Spartan-3. The BCLK signal must be a stable 24 MHz, so it cannot be generated in software. The interface uses four lines: BCLK, SYNC, SDO and SDI, and a frame of 1000 bits gives 48 kHz; for 96 and 192 kHz the number of samples per frame increases.
The implementation uses four PIO state machines: SDO output, SYNC output, SDI input with BCLK generation, and a state machine that assigns an address to the codec. Data moves over DMA, while the RP2040's second core packs samples and commands and parses the codec's replies. The demo project generates sine waves on five stereo DACs, with the third harmonic around −123 dB.
The codec ran at a 192 kHz sample rate. Overclocking the RP2040 to 32 MHz (256 kHz) worked, but the codec heated up considerably. The board draws 150 to 270 mA at 3.3 V. There are limits too: the built-in filters act as a first-order high-pass with a cutoff of about 4 Hz, and the HDA input bandwidth does not allow three stereo channels at 24/192 simultaneously.
The author also found a synchronization bug in the RP2040 when DMA, USB and XIP sampling run together, whose cause was identified by specialists at Raspberry Pi. For complex processing he recommends the RP2350, which has DSP instructions. I like that the project honestly shows both its results and its weak points.
Source: Habr
Sensors have stopped being the almost invisible components of a workshop or factory. Hardware Libre surveys several projects in Spain and Europe where networks of connected devices, data platforms and artificial intelligence are used for very different tasks: anticipating breakdowns, monitoring pollution, identifying birds by their song, or improving vehicle safety.
The most detailed example is Scamia in Madrid, a project of the municipal housing company EMVS that will be applied to fifteen buildings by 2027, thirteen of them public housing. It has a budget of about 5.2 million euros with European co-financing and revolves around a digital twin of each building. Data from roughly two hundred sensors per building, connected via Wi-Fi and IoT, is analysed to measure temperature, relative humidity or CO2 levels and detect anomalies before the neighbours notice them. The data is anonymised and linked to households, not to individuals.
The people in charge estimate a 15% to 20% reduction in maintenance costs for equipment such as aerothermal systems or boilers, and a 5% to 10% cut in energy consumption. The article adds a dose of realism: those percentages come to roughly 40 or 100 euros a year per home in energy, and the real value shows up when you add up many homes and avoid costly breakdowns.
In Cartagena another network is being prepared for the northern neighbourhoods, with a contract of 130,258 euros including VAT and around fifteen devices that will measure gases such as nitrogen dioxide or ozone.
I think it is a good reminder that the same principles we use in a home project, with a microcontroller and a temperature sensor, scale up to a whole city. The hard part, as the text points out, is not measuring but turning millions of readings into useful decisions.
Source: Hardware Libre
Pavel Loktev published a review on Habr after living with the Elecrow CrowPanel wall panel on ESP32-P4 for a week, and looking into why such a device is needed at all. I'm passing on the main points, because I appreciated how honest his conclusions were.
The device has a 10.1-inch 1024×600 screen, 32 MB PSRAM, 16 MB flash, a capacitive touch sensor, audio, an SD slot and GPIO. The firmware is closed, so this is a ready-made product rather than a development platform. The panel works as a client: you install an add-on on the Home Assistant server, then enter the server address, login and password on the panel itself. After that it shows a regular HA dashboard, and there is no need to design a separate interface.
The author provides measurements. A cold start to a working dashboard takes about 20 seconds. Pressing a button until its state changes takes 0.7–0.8 seconds, and switching a tab until the block is ready takes 1.1–1.3 seconds. The command goes out almost immediately, but the visual confirmation on the screen comes noticeably later. Sliders work fine, and he also tested cameras and media playback.
He is honest about the alternatives. A tablet is faster and more versatile, but it needs a mount, power and kiosk mode. ESPHome with LVGL is cheaper, but you have to build and maintain the interface yourself. Ready-made panels such as the Sonoff NSPanel solve the housing and power issues, but often require third-party firmware. CrowPanel is a happy medium: one dashboard for phone, PC and wall.
The author lists lower performance, a resolution that is too small for dense dashboards, and closed firmware as drawbacks. There is no built-in battery. The conclusion: this is not a replacement for a tablet, but a replacement for a control remote at the entrance, in the kitchen or in the garage. I liked his phrasing about a switch with a large screen. I think it's very accurate.
Source: Habr
