Distributional Random Network Distillation (DRND)

Code for ICML 2024 paper "Exploration and Anti-exploration with Distributional Random Network Distillation".

1. Setup

To begin, create a conda environment and activate it using the following commands:

conda env create -f environment.yaml
conda activate drnd

2. Training

2.1 Running Offline Experiments

Quick start by running the following code:

cd offline
sh train.sh

If you need to run other datasets with different hyperparameters, here is an example:

sh train.sh --env_name walker2d --dataset_name walker2d_medium  --actor_lambda 10.0 --critic_lambda 10.0

2.2 Running Online Experiments

If you want to run Atari Game environments, run:

cd online
python train.py

Citation

@article{yang2024exploration,
  title={Exploration and Anti-Exploration with Distributional Random Network Distillation},
  author={Yang, Kai and Tao, Jian and Lyu, Jiafei and Li, Xiu},
  journal={arXiv preprint arXiv:2401.09750},
  year={2024}
}