Thank you for visiting our repository for "BootStrapped Reward Shaping" (BSRS)

Run experiments based on globally set variables:

Specify the experiment parameters in the config.py file. (No arguments are passed to the experiment.py file) The wandb config files for each algo are in the sweep_configs folder.

python experiment.py

If you would like to cite our work, please use the following bibtex based on the AAAI proceedings:

@inproceedings{adamczyk2025bootstrapped,
  title={Bootstrapped Reward Shaping},
  author={Adamczyk, Jacob and Makarenko, Volodymyr and Tiomkin, Stas and Kulkarni, Rahul V},
  booktitle={Proceedings of the AAAI Conference on Artificial Intelligence},
  volume={39},
  number={15},
  pages={15302--15310},
  year={2025}
}

Contributions

Contributions are always welcome! Feel free to add your own algorithm with bootstrapped reward shaping. The general idea applies to any RL algorithm that has access to V(s) the state-value function, for calculating the potential function.

TODO:

• Add pip / poetry installable setup

Future Work / Ideas:

• (Learned) schedule for eta as training progresses
• eta(s) - state-dependent theory holds
• Improve the bounds