Arduino-Vision

Problem

Vision is assumed to need a big computer. The question behind Arduino-Vision is the frugal one: can a microcontroller carry vision? What does machine learning look like when the entire computer is a few kilobytes of RAM and a clock measured in megahertz?

Difficulty

The difficulty is the constraint. A microcontroller has no GPU, no operating system to hide behind, and almost no memory. Every part of the vision pipeline that a laptop would absorb has to be rebuilt within the budget — or dropped.

Built

Arduino-Vision — a combination of ML and Arduino. A pre-trained YOLOv5 model detects objects; the Arduino drives directional LED feedback (pins 9/6/11/6) and a distance buzzer (pin 5), with a push button to reset the ideal state (pin 2). A prototype that puts a machine-learning model on a microcontroller and asks it to see.

Owned

The whole experiment — the model, the Arduino integration, and the decision about what “vision” is allowed to mean at that scale.

Failed

The prototype is a prototype because the gap between what a microcontroller can do and what vision wants is real. The failures were the point: they map exactly where the budget runs out.

Changed

It changed what I expect from embedded perception. Forcing vision onto a microcontroller makes the cost of every operation visible in a way a laptop never does.

Machine-now

A prototype — the combination of ML and Arduino exists and runs, and it stands as the record of the frugal/architectural-curiosity instinct: what is the smallest machine that can see?

Lesson

The constraint is the teacher. Putting vision on a microcontroller shows you exactly what vision costs — and what it costs is usually more than the abstraction lets you see.