Street Fighter III Agent

Developing a deep reinforcement learning agent to play Street Fighter III.

Acknowledgements

This is a project for CSC 570; Aritificial Intelligence in Games, taught by Prof. Canaan at Cal Poly San Luis Obispo.

We reply heavily on the following libraries:

• Diambra
• Stable-Baselines3
• Stable-Baselines3-Contrib

Authors

• Abigayle Mercer (abmercer@calpoly.edu)
• Braedan Kennedy (bkenne07@calpoly.edu)
• Damian Dhesi (ddhesi@calpoly.edu)
• Sarah Duncan (sdunca07@calpoly.edu)
• Yayun Tan (ytan15@calpoly.edu)

Setup

git clone git@github.com:SDuncan5/1v1-Fighting-Game.git
cd 1v1-Fighting-Game
./setup.sh

Training

To configure training parameters:

vim cfg_files/sfiii3n/sr6_128x4_das_nc.yaml

To start training with 8 parallel environments:

sudo service docker start
source .venv/bin/activate
diambra run -s 8 -r "$PWD/roms/" python training.py --cfgFile "$PWD/cfg_files/sfiii3n/sr6_128x4_das_nc.yaml"

Evaluating

To start evaluation of an agent:

sudo service docker start
source .venv/bin/activate
diambra run -r "$PWD/roms/" python evaluate.py --cfgFile "$PWD/cfg_files/sfiii3n/sr6_128x4_das_nc.yaml"

Tensorboard

source .venv/bin/activate
tensorboard --logdir results/sfiii3n/sr6_128x4_das_nc/tb