Reinforcement Learning with Dual-Observation

This is the implementation of the paper "Reinforcement Learning with Dual-Observation for General Video Game Playing" accepted by the IEEE Transactions on Games in 2022. The basic reinfocement learning algorithm is adapted from stable-baseline3.

Please use this bibtex if you use this repository in your work:

@article{hu2022reinforcement,
  title={Reinforcement Learning with Dual-Observation for General Video Game Playing},
  author={Hu, Chengpeng and Wang, Ziqi and Shu, Tianye and Tong, Hao and Togelius, Julian and Yao, Xin and Liu, Jialin},
  journal={IEEE Transactions on Games},
  pages={accepted},
  year={2022},
  publisher={IEEE}
}

Check Generic Video Game Competition (GVGAI) Learning framework from game environment.

Requirments:

pip install -r requirements.txt

Run

Follow the basci running command. Refer to ./arguments.py for more options

python train.py --algo PPO --total-timesteps 1000000 --env-name golddigger

Note:

Please modify following variables according you own setting in ./environment/GOLOEnv.py.

da_environment_path 
gvgai_path