/DoomRL

Training Deep RL agents in VizDoom.

Primary LanguagePython

DoomRL

To run code, set the architecture and config_file variables in train.py and run the command:

python train.py

Acceptable values for architecture are:

  • drqn
  • c51
  • a2c

Model weights are saved in the models directory and the screen buffers of every thousandth game are saved in the gif directory.

The agents directory consists of modified code from VizDoom-Keras-RL while the scenarios folder contains the doom game maps and configurations.