AIoT: Intelligence on Low Power Microcontroller, with MicroPython

By Andri Yadi

Elevator Pitch

AI + IoT = AIoT is the latest advancement of AI and IoT. Let’s move beyond buzzword and seeing it in action. This session will show how to use MicroPython superpower to make Machine Learning (ML) inference runs right on the low power Microcontroller (MCU), not in the cloud, to build AIoT application


Running Machine Learning (ML) inference on GPU-powered full-pledge computer or in the cloud to build AI-powered application is so yesterday. The current trend is Edge Intelligence which emphasises on running AI/ML inference right at the Edge, or user-facing devices, that opens new field of AI and IoT convergence, hence AIoT.

Edge Intelligence is not only allowing much better user experience, thanks to no latency of waiting ML inference result if done in the cloud, but it also improves user privacy as no data ever need to leave the device for ML inference. Plus some bonus, no internet connection is needed, and power consumption is more efficient as one of the biggest power hungry parts in IoT device is connectivity.

Over 9 billion of Microcontroller (MCU) powered devices are built and deployed every year. For perspective, that’s more devices shipping every single year than the world’s entire human population. Possibility to exercise AI on those devices will open a whole new use cases and business opportunities, and solve more problems. While a few of these devices are capable of running ML inference today, but within just a few years, this entire industry, all 9 billion or more devices per year, is on path to Edge Intelligence. By the advancement of ML algorithms and tooling, and accompanied by the cheaper, more powerful MCU with built-in AI accelerator, Edge Intelligence is inevitable.

AIoT or Edge Intelligence seems still far off, but it’s not, we can exercise it today. In this session, I’ll explain and demo to the audience that it’s totally possible.

I’ll show how to:

  • Train Machine Learning model for some use cases: object classification, object detection, anomaly detection.
  • Prepare that ML model so it can be used for inference on MCU.
  • Run the inference on MCU. Best of all, using MicroPython

I’ll also discuss about the hardware and software platform options to build AIoT-powered applications.

Let’s be part of future problem solving - powered by AIoT - by joining my session.