ENGLISH | 中文

This is the implementation of An Imitative Reinforcement Learning Framework for Pursuit-Lock-Launch Missions. The expert dataset, the trained models, and recorded videos of the learned policies are available at Google Drive.

Please follow the official instructions to install Dogfight Sandbox.

conda create -n hirl python=3.8
conda activate hirl
git clone https://github.com/zrc0622/HIRL4UCAV.git
cd HIRL4UCAV
conda install pytorch==1.13.0 torchvision==0.14.0 torchaudio==0.13.0 pytorch-cuda=11.7 -c pytorch -c nvidia
pip install -e .

Download the expert_data and bc_actor folders from Google Drive and place them in the repository, then update the IP address in the local_config.yaml file.

Once the prerequisites are set, follow these steps to run experiments:

1. Start the Environment: Run the Dogfight Sandbox and choose the NETWORK MODE mission

2. Start Training: With the environment running, execute the following command to run the experiments:

   python hirl/train_all.py --port=<ENV_PORT> --env=<ENV_TYPE> --random --agent=HIRL --type=<HIRL_TYPE> --model_name=<MODEL_NAME>

Replace placeholders with your specific setup.

@misc{li2025imitativereinforcementlearningframework,
      title={An Imitative Reinforcement Learning Framework for Pursuit-Lock-Launch Missions}, 
      author={Siyuan Li and Rongchang Zuo and Bofei Liu and Yaoyu He and Peng Liu and Yingnan Zhao},
      year={2025},
      eprint={2406.11562},
      archivePrefix={arXiv},
      primaryClass={cs.LG},
      url={https://arxiv.org/abs/2406.11562}, 
}