Use the deep reinforcement learning algorithm DQN to play the game SuperMario.
The game environment is come from https://github.com/justinmeister/Mario-Level-1.git
The agent performance is not good enough.Actually there is a gap between my expectations and the fact.This kind of learning algorithms are magical and hard to control.There must be some internal logic between the agent performance and a bunch of hyperparameters.And it is important to note that the supermariobros game is complicated than Atari games in gym,the game frame is more rich and has more changes.
Anyway,you can just clone this repository and run "play_mario.py"
- python3.5
- opencv-python
you can install it using conda
conda install -c menpo opencv
- pytorch
- the supermario game environment