[Editor's Note] The Arduino UNO Q is a hybrid Edge AI board combining the Qualcomm QRB2210 MPU and STM32 MCU, supporting both on-device AI and real-time control. It provides GPU and DSP acceleration and a Linux and Docker-based development environment, making it optimized for use in industrial AI fields such as smart factories, predictive maintenance, security, and robotics. Accordingly, we met with Carelab CEO Bong-jo Kim, who will be hosting a webinar on July 30 titled 'A New Chapter for Arduino: Edge AI Development Starting with Arduino UNO Q,' to hear more about edge AI development using Arduino UNO Q.


■ I would like to know what the Arduino UNO Q is, its hardware architecture, and its technical differences.

The Arduino UNO Q is a monumental all-in-one prototyping board released by Arduino after partnering with Qualcomm.

Completely departing from the conventional concept of simple 8-bit/32-bit MCU boards, it adopts a hybrid architecture that combines a high-performance application processor (MPU) and a microcontroller (MCU) for real-time control on a single board.

The MPU (Application Processor) is the Qualcomm® Dragonwing™ QRB2210, and ISP & AI acceleration supports machine vision and edge AI computations through the inclusion of a Dual ISP and an integrated AI accelerator.

The STMicroelectronics STM32U585 MCU (real-time microcontroller) is capable of handling precise and immediate real-time tasks, such as sensor data acquisition and motor control, as well as low-power system management.

■ How the inclusion of high-performance computing elements such as Adreno GPUs, VPUs, and DSPs changes Edge AI development I'm curious

High-performance computing and multimedia elements such as the Adreno GPU, Hexagon DSP, and Dual ISP (Image Signal Processor) integrated into the Qualcomm QRB2210 processor completely change the development paradigm of existing MCU-based development or general CPU-centric SBCs (such as the Raspberry Pi).

Maximizing computational efficiency through heterogeneous computing, a dramatic increase in frames per second (FPS) and reduction in latency, innovation in thermal management and power efficiency, and the implementation of multimodal and complex edge applications are possible.

The change developers will experience is the ability to configure a true 'on-device multimodal AI stack' that handles sophisticated scenarios—such as waking up by listening to the user's voice (DSP), identifying people via the camera (ISP), and responding by analyzing facial expressions (GPU)—exclusively on the device without external cloud connectivity.

■ I am curious why VS Code and SSH-based development was officially supported. In addition, the development environment combinations recommended for beginner and advanced developers, respectively, are

The reason the Arduino UNO Q officially supports (recommends) development workflows based on VS Code and Remote-SSH is due to the unique characteristics of the board's 'hybrid (MPU + MCU) architecture'.

The reasons for officially supporting VS Code and SSH-based development include the heterogeneous collaboration structure between Linux (MPU) and real-time control (MCU), the absorption of embedded AI/deep learning development workflows, and organic integration with CLI (Command Line Interface).

For beginner developers who are just starting out with hybrid boards or aiming for rapid prototyping, we recommend the single combination of Arduino App Lab (Arduino App Lab + built-in terminal), a dedicated tool introduced by Arduino alongside the UNO Q.

For professional developers who design system architectures directly and are focused on on-device AI model acceleration and productization, we recommend a powerful, fully customizable professional toolchain (VS Code + Remote-SSH + Arduino App-CLI).

■ I am curious about the representative use cases for Edge AI that UNO Q aims for. I would also like to know the level of on-device inference performance based on the QRB2210 and which models are most suitable.

Thanks to an architecture that combines high-performance Linux computing (MPU) and real-time low-power control (MCU), UNO Q is optimized for complex scenarios where 'AI makes decisions and hardware responds immediately,' going beyond simply analyzing data (Smart Factory and Industrial Predictive Maintenance, Smart Retail and Vending, Robotics and Intelligent AGVs).

While the Qualcomm QRB2210 aims to be an ultra-compact, low-power embedded platform, it incorporates hardware accelerators (Adreno GPU and Hexagon DSP), delivering performance on a completely different level compared to general CPU-based SBCs.

Considering the board's memory (2GB to 4GB LPDDR4) and computing power of 1 to 2 TOPS, mobile models that are lightweight (Quantized, INT8/FP16) for edge environments yield the best efficiency, rather than large models with too many parameters.

To extract the best performance from the UNO Q, it is essential to perform a wrangling process to quantize the model into 8-bit (INT8) using Qualcomm's SNPE (Qualcomm Neural Processing SDK) or TensorFlow Lite (with GPU/Hexagon Delegate) rather than running the model as is. This process allows you to utilize the accelerator's potential 100% while drastically reducing memory usage.

■ UNO Q features a structure suitable not only for manufacturers but also for industrial Edge AI development. Which sectors in the industrial field are expected to adopt it the fastest?

The biggest strength of the Arduino UNO Q is that 'Linux-based high-performance AI computing (Qualcomm MPU)' and 'latency-free physical control and 3.3V/5V sensor interface (STM32 MCU)' are perfectly integrated on a single board.

In existing industrial settings, a complex configuration involving connecting a 'Raspberry Pi (or industrial PC) + Arduino' with communication lines had to be adopted to implement these two things, which led to chronic hardware risks such as frequent connector failures, communication noise, and increased bulk.

Considering this integrated architecture and support for Docker-based container orchestration, the three key areas where this board will be adopted most rapidly in industrial settings are: 'vision-based AI inspection and smart grippers' (machine vision + motor control) in smart factories; 'on-device biometrics and PPE inspection' (security + field independence) for enterprise security and safety management; and 'always-on-the-go predictive maintenance edge nodes' (low-power DSP + real-time sensing) for equipment asset management.

The biggest appeal for field developers is that they can easily port a Linux OS-based Docker distribution system and Qualcomm Edge AI runtime to smart factory infrastructure (PLC, CAN, MQTT, etc.) simply by replacing the board, while maintaining the existing Arduino UNO’s vast shield ecosystem and wiring structure.

It will shorten the time from on-site POC (Proof of Concept) to mass production by more than several months.

■ Along with an introduction to the webinar, do you have any advice for developers using UNO Q?

Ahead of this exciting journey, I would like to offer some technical advice to developers who are considering taking UNO Q to the next level of productization and field application.

First, it is requested that the 'Aggregation of Roles (AMP)' structure of the two processors be designed. The most frequent mistake in UNO Q development is blurring the boundary between the MPU (Qualcomm Linux) and the MCU (STM32).

Next, I urge you not to hesitate to 'quantize' to take advantage of accelerators (GPU/DSP). Although the RB2210 processor houses a powerful Adreno GPU and Hexagon DSP, if you run standard FP32 (32-bit floating-point) models as they are, only the general-purpose CPU will suffer.

We must actively adopt a Docker-based container deployment environment. Since the UNO Q runs on Debian Linux, field maintenance and deployment architecture design are highly flexible.

It is recommended to use it as a bridge from Proof of Concept (PoC) to mass production. The biggest advantage of the Arduino UNO is its vast shield ecosystem and intuitive breadboard wiring. The UNO Q inherits this form factor exactly.

Arduino is no longer just a coding educational tool for children. Combined with Qualcomm's powerful physical AI architecture, the UNO Q is a powerful weapon capable of instantly transforming an engineer's imagination into industrial terminals.

I hope that through this e4ds webinar, you will clearly recognize the hardware limitations and possibilities of the board, and furthermore, gain valuable hints to break down the persistent barrier of “Why does the product work when the examples do?”