An unfinished version of self-driving AI in GTA5
Start from the ipython notebook. If you wanna know the requirement of packages, see the import commands in each block.
This is an simplified version of Charles self-driving AI:
1.less actions: W A D WA WD NK.
2.no obstacles, main goal is just stay in lines while driving.
3.no dashboard, full screen of scene.
However, due to shortage of storage, time and data batches, I haven't finished the train process of this AI.
My train data was not enough, so the loss value just won't decrease, and the damn accuracy just stay near 0.73 and wouldn't go up even for only a little bit!!!!
I just wonder if there is any method or any other model than desnext can reduce the scale of required data. lol
Awesome dudes, I need ur help :(
————JackieFangTheBorg