On the computer vision it seems that advanced navigation tasks (DroNet) have been run completely onboard on very little processors (e.g. GAP8 – Ultra low power 9 core RISC-V MCU) like the one from the PULP platform with a Himax HM01B0 camera (Ultra low power 320×240 grayscale @ 60fps)
GAP8 seems to have a built in tool to convert neural networks trained in python to their embedded platform. Do you think it is too much underpowered?
I’ve also seen usage of FPV cameras for AI research with offboard control on ROS.
What about JeVois and OpenMV?
By the way the K210 seems a very good option…