Looks like the motor controller is working fine, it’s the Jetson Nano’s GPIO header that’s in backwards...

So I could use a spare Pi 3B+ (assuming I get though the ROS installation song & dance), but I'd lose all the cool GPU capabilities. I have a neural compute stick, but that's really only for vision, and I'd like to experiment with other neural net options, too (e.g. using translational embedding to build sort a neural-network-as-a-memory). I might try reconfiguring the hardware to see if I can get in a better layout. We'll see.