In this repo, I used some math and image manipulation skills to create my own reinforcement learning environnement for autonomous car
this environnement is simple: a race track with white borders, in this environnement, a car (represented by a red point) is evolving, his goal is too survive as long as possible (so make as many laps as possible)
rewards are :
- if the car is going forward then reward = current speed
- if the car is turning then reward = half the current speed (to prevent from turning too much)
- if the car is making an half turn reward = - 45
- if the car is going out of the track reward = - 150 and break the run/ cause respawn + AI training
the state image is with perspective, you can tweak it in the 3D_map function in env.py I made some function to add some texture, light effects to be more realistic and challenging for the AI