Can Arduino be used in machine learning applications?

Arduino can be used for machine learning applications, but with some important limitations. Classic boards such as the Arduino Uno or Nano are not suitable for advanced computing, but modern models with 32-bit microcontrollers allow the implementation of simplified AI models within a domain known as TinyML.

What is TinyML and how does it relate to Arduino?

TinyML is a field of engineering that deals with implementing machine learning algorithms on devices with very limited resources, such as microcontrollers. Instead of processing data in the cloud or on a computer, the analysis and decision-making is done locally - on the device itself. This means lower latency, no need for an internet connection and less power consumption. Arduino fits into this trend with models such as the Arduino Nano 33 BLE Sense, which are based on a more powerful ARM Cortex-M4 processor and have additional memory resources. In contrast to the classic ATmega328P (2 KB RAM), this model offers up to tens of times more memory and built-in sensors (e.g. microphone, accelerometer, temperature sensor), allowing local processing of sensor data.

What AI applications are possible on the Arduino?

On TinyML-compatible platforms, it is possible to run small neural network models, previously trained on a computer and optimised to run on a microcontroller. Examples of such applications include: -gesture recognition based on accelerometer data, -sound analysis for detecting keywords or unusual sounds, -detection of anomalies in measurement data (e.g. machine vibrations). These models are usually created in the TensorFlow Lite environment and then exported to a format that can be implemented in Arduino code. Dedicated libraries, such as Arduino TensorFlowLite, are used for this purpose.

Boundaries and technical requirements

Be aware that the classic Arduino Uno or Mega boards are not suitable for running even basic machine learning models. Their memory resources (2KB RAM, 16MHz clocking) are not sufficient to support the TinyML library. For AI applications, it is necessary to use 32-bit boards, such as the aforementioned Nano 33 BLE Sense, ESP32 or STM32, which have much higher computational and memory capabilities. Another limitation is that the Arduino does not learn on its own. The model has to be pre-learned on the computer and only the ready-made, optimised decision code goes to the microcontroller. However, this is sufficient for most embedded applications where fast response and lack of network dependency are crucial.

Transfer Multisort Elektronik (TME) is one of the world’s largest global distributors of electronic components, electrotechnical parts, workshop equipment, and industrial automation. The catalog includes over 1,300,000 products from 1,300 leading manufacturers. TME’s modern logistics centers in Łódź and Rzgów (Poland), with a combined area of over 40,000 m², ship nearly 6,000 packages daily to customers in more than 150 countries.

TME also invests in the development of knowledge and skills of young engineers and electronics enthusiasts through the TME Education project, and supports the tech community by organizing the TechMasterEvent series, promoting innovation and experience exchange.