In these lessons you will develop a Machine Learning model using the YOLO algorithm to detect FRC game pieces. The model will then be deployed on the Raspberry Pi using the Coral USB Edge TPU and connected with NetworkTables to stream the inference results to the robot and driver station.

As a note for posterity, other mobile ML chips were also evaluated for this project but were not used:

• Raspberry Pi AI Camera: This is a slick device with the camera and ML chip on a single device, and getting it running on the Raspberry Pi was the easiest out of the options evaluated. However, the limitations of having the camera co-located with the Pi (connected by a ribbon cable) and the inability to switch to different cameras like a global shutter camera ruled this option out.

• Raspberry Pi AI Hat: This uses a high performance HAILO chip to do ML on the Raspberry Pi. This would be the fastest implementation and would allow any camera to be used. Unfortunately, the ecosystem around the AI Hat, such as documentation and detailed example code, just isn't there yet. In particular the process of compiling a ML model into the right format for the hat was just too cumbersome. Hopefully this will improve for next season.