Robot Car was an early byitl project — a tabletop-scale autonomous rover built to learn what it takes to put real-time computer vision on a moving platform, without dragging a laptop around behind it.
The build
The first iteration was three wheels: two rear driven wheels, a caster up front, a Raspberry Pi with the standard camera module rubber-banded to the top, and a pair of yellow gear motors running off a basic H-bridge. Frames streamed back to a laptop over Wi-Fi for OpenCV work — object tracking, line following, that class of thing. Simple, but it was our first "the code sees what the world is doing" moment.
The second iteration added a Google Coral USB Accelerator alongside the Pi and traded the caster for a proper four-wheel drivetrain. With the Edge TPU handling inference on-device, the rover no longer needed the streaming-to-laptop crutch — it ran classification and detection at real-time frame rates completely stand-alone, which was the whole point of building it.
What we learned
Two things. First: getting the mechanicals right — motor mounts, wheel alignment, camera stiffness — took as long as the vision code. A flexy 3D-printed part will kill your detection consistency long before your model does. Second: the Coral was excellent for its era, but the interesting work — motor drivers, PCB fit, chassis serviceability — always came back to enclosure design.
That pattern kept repeating across our early prototypes, which is how byitl ended up as a PCB-to-enclosure service today. Robot Car was one of the builds that made the case.