vietnh1009/Super-mario-bros-A3C-pytorch

Implements multi-worker parallel training across independent environment instances with asynchronous gradient updates to a shared model, leveraging PyTorch for the actor-critic network architecture. Handles raw pixel input through CNN feature extraction and trains agents across multiple Super Mario Bros stages without complex preprocessing pipelines. Includes pre-trained model weights and supports both training via `train.py` and inference via `test.py` with standard gym environment integration.

1,108 stars. No commits in the last 6 months.