Learning to Solve Tasks with Exploring Prior Behaviours

This repository contains the official implementation for IRDEC introduced in the following paper:

Learning to Solve Tasks with Exploring Prior Behaviours

Ruiqi Zhu, Siyuan Li, Tianhong Dai, Chongjie Zhang, Oya Celiktutan

Environment

• Python 3.7
• Other required python packages are specified by requirements.txt.

Prerequisites

• Install Mujoco and other required packages.
• Register Wandb account.
• Clone this repository.

Troubleshooting

if meet RuntimeError: CUDA error: CUBLAS_STATUS_EXECUTION_FAILED, which is caused due to the cuda toolkit.

pip3 install torch==1.11.0+cu113 -f https://download.pytorch.org/whl/cu113/torch_stable.html

Reproducing Ant-Umaze Experiment

chmod +x ant_umaze_train.sh
./ant_umaze_train.sh

Reproducing Ant-FourRoom Experiment

chmod +x ant_four_room_train.sh
./ant_four_room_train.sh

Citation

If you find this code useful in your research, please consider citing:

@article{zhu2023learning,
  title={Learning to Solve Tasks with Exploring Prior Behaviours},
  author={Zhu, Ruiqi and Li, Siyuan and Dai, Tianhong and Zhang, Chongjie and Celiktutan, Oya},
  journal={arXiv preprint arXiv:2307.02889},
  year={2023}
}