This project features a deep reinforcement learning model training an Actor-Critic agent for the Pommerman game AI competition. The game itself is given to us via the competition.
Check out our youtube video for how the model performs in action.
Go to [https://github.com/MultiAgentLearning/playground/tree/master/docs] and follow the guide there on installing the Pommerman game. You can use the game folder from this repository for installing.
There is a jupyter notebook file in the src folder for running the code. From here it is possible to both train a new model and visualize a pretrained one. It takes around 20 hours of training on a decent GPU to see real progress.