Official PyTorch implementation of the paper "Value-Distributional Model-Based Reinforcement Learning".

Prerequisites:

• conda (optional, install for Option #1 below)
• docker (optional, install for Option #2 below)

1. Clone the repository and cd into it

git clone https://github.com/boschresearch/dist-mbrl.git && cd dist-mbrl

2. Create a conda environment

conda env create --file=environment.yml

3. Activate the environment and install the package + dependencies

conda activate dist_mbrl
pip install -e .

Make sure docker is installed and configured.

1. Build docker image

cd docker/
./build_docker.sh

2. After the image is created, you can access it via

docker run --rm -ti dist-mbrl

The entry point for training is train.py. At the bottom of the file, you can modify the configuration passed on to the training script. The agent configurations are generated in default.py and the model learning configurations are stored in this YAML file.

cd {path_to_repo}/dist_mbrl
python train/train.py

All the plots shown in the paper can be reproduced by running the provided Jupyter notebooks.

@article{luis2023value,
  title={Value-Distributional Model-Based Reinforcement Learning},
  author={Luis, Carlos E and Bottero, Alessandro G and Vinogradska, Julia and Berkenkamp, Felix and Peters, Jan},
  journal={arXiv preprint arXiv:2308.06590},
  year={2023}
}

The code is open-sourced under the AGPL-3.0 license. See the LICENSE file for details.