Skip to content

The official material of the paper: "Improving the Generalization of End-to-End Driving through Procedural Generation".

License

Notifications You must be signed in to change notification settings

decisionforce/pgdrive-generalization-paper

Repository files navigation

Status: Archive - code is provided as-is, no updates expected.

Improving the Generalization of End-to-End Driving through Procedural Generation

This is the official material of the paper: "Improving the Generalization of End-to-End Driving through Procedural Generation".

Please visit the following links to learn more on our PGDrive simulator:

[ 📺 Website | 🏗 PGDrive Repo | 📜 Documentation | 🎓 Paper ]

Setup the environment

# Clone this repo to local
git clone https://github.com/decisionforce/pgdrive-generalization-paper.git
cd pgdrive-generalization-paper

# Install dependencies
pip install -r requirements.txt

Draw the main experiments' results

Before you draw the results, please decompress the /data.zip to /data folder.

# Generate result to results/ppo-main-result-up.pdf and results/ppo-main-result-down.pdf
python draw_ppo_results.py

# Generate result to results/sac-main-result-up.pdf and results/sac-main-result-down.pdf
python draw_sac_results.py

Draw the other experiments' results

# Draw change density experiment result to results/change-friction-result.pdf
python eval_density.py

# Draw change friction experiment result to results/change-friction-result.pdf
python eval_friction.py

Reproduce the experiments

# PPO main experiments
python train_ppo.py --exp-name main_ppo --num-gpus 0

# SAC Main experiments
python train_ppo.py --exp-name main_ppo --num-gpus 0

# Change friction experiments
python train_ppo_change_friction.py --exp-name change_friction --num-gpus 0

# Change density experiments
python train_ppo_change_density.py --exp-name change_density --num-gpus 0

# For the change density / friction experiments, you need to uncomment the script to call
# get_result function, then call
python eval_density.py
python eval_friction.py

Cite this work

If you find this work useful in your project, please consider to cite it through:

@article{li2020improving,
  title={Improving the Generalization of End-to-End Driving through Procedural Generation},
  author={Li, Quanyi and Peng, Zhenghao and Zhang, Qihang and Qiu, Cong and Liu, Chunxiao and Zhou, Bolei},
  journal={arXiv preprint arXiv:2012.13681},
  year={2020}
}

About

The official material of the paper: "Improving the Generalization of End-to-End Driving through Procedural Generation".

Resources

License

Stars

Watchers

Forks

Languages