I started with reinforcement learning a relatively short time ago and tried to program several algorithms myself. My current project was to write the Soft Actor-Critic algorithm and then use it for autonomous driving in Carla.
My code manages to solve simple tasks, but when I try it on Carla, I still achieve poor performance, even though I base my rewards on master’s theses and these achieve better performance. Do I have a programming error somewhere in my code or a mathematical error that it does not learn. It would be very kind if someone would look at my code and give feedback.
My Code is on Github with following link: b-gtr/Soft-Actor-Critic: Soft-Actor-Critic with image preprocessing