We analyze the effect of different types of image pre processing techniques on training an RL agent. The task is to learn to Flappy Bird using Deep Q-Learning. You can find more details about the experiments in our report [link to report].

conda create -n birdRL pytorch==1.0.1 torchvision==0.2.2 cudatoolkit=10.0 -c pytorch
conda activate birdRL
conda install tensorboardX
pip install pygame==1.9.4 opencv-python==3.4.4.19

The folders Binary, Flow, etc. are the experiments with different image pre-processing. cd to any folder and run the following command.

cd Binary
python train.py

You can check the tensorboard logs using:

tensorboard --logdir .

Similar to training:

cd Binary
python test.py