Training a Doom-playing agent using Deep Q-Networks (DQN). This project provides:
-
A custom VizDoom environment with shaped rewards
-
A PyTorch DQN implementation (CNN encoder + MLP head)
-
A training script that logs reward components per episode
-
Evaluation and gameplay visualization (replays, GIFs)
-
ONNX export of the trained policy for seamless deployment
-
Clone, install dependencies, and run train.py to get started!