Really helpful post! Thank you.
Django apps usually also use nginx as the reverse proxy and to serve static files. Is nginx not required when using Google Cloud’s Flexible app engine?
In my case I was using the GlobalPlanner from the ROS Navigation Stack. The update rate was set to 5 times a second. This was good enough to detect slow moving or fixed dynamic obstacles (i.e. obstacles not present in the map).
I am not using the Lambda layer to do the normalization. There is a bunch of pre-processing, including normalization, happening before the data is fed to the model. Have a look at the code in the github repo. There is a function ‘get_augmented_row’.
I bought the VESC from a Chinese company Maytech, through aliexpress. The hardware version is 4.12 and firmware version was 3.38. The MIT Racecar vesc repository has code for firmware 2.18, so you need to make some changes to it if you are using firmware 3.x. See this https://github.com/mit-racecar/vesc/pull/13
You could use a brushed motor as well. I had thought about it as well. Brushed motors are cheaper. However the car model, which I bought, had a brushless motor on it. My choice was constrained by the available car models in my city.