This video shows the automated docking with the ISS of the SpaceX's Dragon Capsule on the SpaceX simulator https://iss-sim.spacex.com/, using a Neural Network PID Controller built using Tensorflow.
This version of the controller use separated controller for each dimension, x, y, z, roll, yaw, and pitch, independently from each other, but the error is projected from the ISS frame of reference to the capsule frame of reference before feeding it to the controller. This allows the controller operate properly under attitude rotation.
The following gist https://gist.github.com/rzilleruelo/f... contains the inference code of the controller.