/ggtde

Generalized Gaussian Temporal Difference Error For Uncertainty-aware Reinforcement Learning

Primary LanguagePython

Generalized Gaussian Temporal Difference Error

Official PyTorch Implementation of Generalized Gaussian Temporal Difference Error For Uncertainty-aware Reinforcement Learning.

Usage

Environment

We manage our packages using poetry. Please refer to the official guide for installation instructions. Once installed, run the command poetry install to install all the required dependencies.

Train

Note: The overall training pipeline follows the structure of RL Baselines3 Zoo.

poetry run python train.py --algo <RL_ALGORITHM>
                           --env <ENVIRONMENT>
                           --trained-path <TRAINED_MODEL_PATH>
                           --seed <SEED>
                           --verbose <VERBOSE_LEVEL>

Arguments:

  • RL_ALGORITHM: The algorithm to use.
  • ENVIRONMENT: The environment to train on.
  • TRAINED_MODEL_PATH: The path to the trained agent.
  • SEED: The random seed.
  • VERBOSE_LEVEL: The verbosity level.

Citation

@misc{kim2024ggtde,
  title={Generalized Gaussian Temporal Difference Error For Uncertainty-aware Reinforcement Learning},
  author={Seyeon Kim and Joonhun Lee and Namhoon Cho and Sungjun Han and Seungeon Baek},
  year={2024}
}