NanoEdge AI Studio for Arduino

Find out more information: http://stm32ai.st.com Transcript: When it comes to artificial intelligence, you might think it's a complex subject reserved for massive servers and high-tech labs. But guess what? We've packed a machine learning algorithm into an Arduino board to recognize hand gestures! So, the power of AI is now at your fingertips, quite literally :). At ST we want everyone to be able to release their creativity and we know that running ML on Arduino is a great way to do so. Here we have an example of an application built with NanoEdge AI studio running on the STM32 of an Arduino GIGA R1 to manage the screen and so on, but we have generated such a small model that you can make it run on every microcontrollers. As you can see when I put my hand in front of the Time of Flight the model is classifying my gesture in real time (the inference time is 35us). Then the Arduino GIGA randomly select a gesture among the possible states and if you’re lucky enough you win the game and you get a gift! let's check out how you can do this in a matter minutes. Let’s start by creating a new project and selecting an Arduino board. Allocate the amount of memory that we want to use and select the sensor.Then we can choose anomaly detection, classification and regression depending the application we want to target. Let's choose a classification and upload our dataset... Et voila ! The studio will find the right AI model and train it locally. You don’t need to be an AI expert, to collect massive amounts of data and to train on big GPUs. This autoML approach is definitely the easiest way to develop AI libraries from scratch! Once the benchmark is over and your AI model is ready. You can download the generated library and integrate it to your project. Of course, if you have chosen an Arduino the library is fully compatible with Arduino IDE. This model can run on any STM32, even on the smallest Cortex M0 based or on the famous Arduino Uno. Indeed, here, the AI model reach more than 99% of balanced accuracy with a size of 130Kb of Flash and less than 1Kb of ram ! By the way, you can easily recreate this application at home by clicking the tutorial link here or here or in description depending editor’s mood. Running AI on the edge provides incredible value for a lot of applications such as predictive maintenance on electrical motor or activity recognition in smart watches to list just a few of them. You can create new exciting features without any additional infrastructure cost. This is one use case among millions so I hope this video gave you some idea of exciting application you can create thanks to AI solutions from ST. You can try the studio from today, find you favorite Arduino board and start building you application for free.

Related AI hardware guides

• edge AI devices