This week, I worked on having an acceleration for the CV by using the laptop instead of the Pi. This allowed us to almost 20x improve the latency. I used imagezmq, a ZMQ, based library to send the images to the laptop. Then, I did processing entirely on a python script, sending the results (command responses) as a string back. This was a little tricky to setup, but not too hard given the library abstracted a lot. This helped a lot with visualization making debugging a lot easier, since we can always see what’s being mispredicted. Looking forward, I need to work on some movement fine tuning so the robot can shake around the fire to get all the angles on it.