Vision Training
Vision Training
From a photon hitting a camera sensor to a pose on the robot, in about 15 milliseconds. A visual tour of our Jetson Orin Nano AprilTag vision system: the hardware, the software and the math, and the engineering we did to make it match-ready.
You don't need to know anything about computers or cameras to start. Each chapter builds on the last, and most have a lab you can play with. Hover a dotted word to see what it means. Part 7 is a reference for when you do it for real: robot code, camera setup and pit fixes.
Why a robot needs eyes, and the special black-and-white squares that make seeing easy.
How a $50 camera turns light into numbers, and the knobs that decide whether a tag is crisp or a smear.
A credit-card-sized supercomputer, the operating system that runs it, and how a picture travels through its memory to 1,024 tiny processors at once.
PhotonVision, the open-source program at the heart of it, and how it runs every camera's pipeline.PhotonVision, the open-source program at the heart of it, and the detective work that took us from 33 to 122 frames per second.
Turning four corners on a screen into "the robot is here, facing that way" — and knowing exactly when "here" was.
Getting answers to the robot controller, and everything we built so the system keeps working when a match gets rough.
For when you're doing it for real: writing the robot's vision code, getting the signs right, setting up a camera, and fixing things in the pit.