Code for "Calibrated Value-Aware Model Learning with Probabilistic Environment Models"

Authors: Claas Voelcker, Anastasiia Pedan, Arash Ahmadian, Romina Abachi, Igor Gilitschenski, Amir-massoud Farahmand

Installation

Please ensure that you have a cuda12 capable GPU installed. Other GPUs can work, but we do not provide installation help. All dependencies are best installed via pip using the provided pyproject.toml and we strongly recommend using uv. With this tool you can simply execute uv run mad_td/main.py and all requirements will be installed in a virtual environment.

The codebase is built heavily on mad-td, so in case of issues, it's always good to double check in that repo as well.

Paper experiments

We provide raw results in the corresponding folder.

Citation

If you use our paper or results, please cite us as

@InProceedings{voelcker2025cvaml,
  title={Calibrated Value-Aware Model Learning with Probabilistic Environment Models},
  author={Voelcker, Claas and Pedan, Anastasiia and Ahmadian, Arash and Abachi, Romina and Gilitschenski, Igor and Farahmand, Amir-massoud},
  booktitle={Proceedings of the International Conference on Machine Learning},
  year={2025}
}