This year, our team is planning to process the image with a jetson tx-1. We are planning to attach the camera usb directly to jetson and transfer all the data from jetson to networktables. But we are suspecting if it is the most efficient way or not. So the question is if it’s the most efficient way or is there a better way to transfer both camera image raw data and processing variables to roborio and pc ?
Keep the processing on the jetson - network latency is your main enemy here. Networktables is totally fine. If you want to get super fancy with it, you could write some custom TCP or use 0MQ or something, but I don’t think that’s quite worth the development overhead.
ZeroMQ is fairly minimal from a development standpoint and I’ve found it to be a bit easier than networktables. That being said, either will work and you are spot on about keeping data internal to the robot as much as possible.