Code for TransDreamer: The code is built upon DreamerV2 codebase. Currently only Atari and Crafter is supported.
- Preparation:
conda env create -f environment.yml
- Train the model:
python main.py --config-file config_files/configs_atari.py
- Results reproduction
- Atari-Boxing
- Atari-Freeway
- Atari-Pong
- Atari-Boxing
@article{chen2022transdreamer,
title={Transdreamer: Reinforcement learning with transformer world models},
author={Chen, Chang and Wu, Yi-Fu and Yoon, Jaesik and Ahn, Sungjin},
journal={arXiv preprint arXiv:2202.09481},
year={2022}
}