PGDrive: an open-ended driving simulator with infinite scenes

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

Welcome to PGDrive! PGDrive is an driving simulator with many key features, including:

• 🎏 Lightweight: Extremely easy to download, install and run in almost all platforms.
• 📷 Realistic: Accurate physics simulation and multiple sensory inputs.
• 🚀 Efficient: Up to 500 simulation step per second and easy to parallel.
• 🗺 Open-ended: Support generating infinite scenes and configuring various traffic, vehicle, and environmental settings.

🛠 Quick Start

Please install PGDrive via:

pip install pgdrive

If you wish to contribute to this project or make some modification, you can clone the latest version of PGDrive locally and install via:

git clone https://github.com/decisionforce/pgdrive.git
cd pgdrive
pip install -e .

You can verify the installation and efficiency of PGDrive via running:

python -m pgdrive.examples.profile_pgdrive

The above script is supposed to be runnable in all places. Note that please do not run the above command in the folder that has a sub-folder called ./pgdrive.

🚕 Examples

Please run the following line to drive the car in the environment manually with keyboard!

python -m pgdrive.examples.enjoy_manual

You can also enjoy a journey carrying out by our professional driver pretrained from reinforcement learning!

python -m pgdrive.examples.enjoy_expert

A fusion of expert and manual controller, where the expect will try to rescue the manually controlled vehicle from danger, can be experienced via:

python -m pgdrive.examples.enjoy_saver

To show the main feature, procedural generation, we provide a script to show BIG:

python -m pgdrive.examples.render_big

Note that the above three scripts can not be run in headless machine. Please refer to the installation guideline in documentation for more information.

Running the following line allows you to draw the generated maps:

python -m pgdrive.examples.draw_maps

To build the environment in python script, you can simply run:

import pgdrive  # Import this package to register the environment!
import gym

env = gym.make("PGDrive-v0", config=dict(use_render=True))
# env = pgdrive.PGDriveEnv(config=dict(environment_num=100))  # Or build environment from class
env.reset()
for i in range(1000):
    obs, reward, done, info = env.step(env.action_space.sample())  # Use random policy
    env.render()
    if done:
        env.reset()
env.close()

We also prepare a Colab which demonstrates some basic usage of PGDrive as follows:

📦 Predefined environment sets

We also define several Gym environment names, so user can start training in the minimalist manner:

import gym
import pgdrive  # Register the environment
env = gym.make("PGDrive-v0")

The following table presents some predefined environment names.

🏫 Documentations

More information about PGDrive can be found in PGDrive Documentation. Besides, the training code of our paper can be found in this repo.

📎 Citation

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}
}