voxelgym2D

A gymnasium environment for voxel/grid based reinforcement learning for path planning.

Results with SB3 (v1.6.2) : PPO 😄

Here are the results of training a PPO agent on the onestep-v0 using the example here. Below you will find the episode reward and episode length over steps during training. As the agent learns, the episode reward increases and the episode length reduces are the agent learns to identify the goal and reach it in the shortest possible path.

Installation

pip install git+https://github.com/harisankar95/voxelgym2D.git

or directly from PyPI 😄

pip install voxelgym2D

Development

To install the package in development mode, run the following command in the root directory of the repository:

git clone https://github.com/harisankar95/voxelgym2D.git ~/path/to/repo
cd ~/path/to/repo
pip install -e .[dev]

# to aditionally install stable_baselines 3 and pytorch (optional)
pip install -e .[dev,sb3]

Usage

import voxelgym2D
import gymnasium as gym

env = gym.make("voxelgym2D:onestep-v0")
observation, info = env.reset(seed=123456)
env.render()

Examples

The examples can be found here.

License

This project is licensed under the terms of the MIT license.

Documentation

The documentation can be found here.

Changelog

0.1.0

• Initial release of voxelgym 2D environments tested with stable_baselines 3 (v1.6.2) and python 3.8

0.2.0

• Available on PyPI, sphinx documentation in work

0.3.0

• Migration to gymnasium
• Agent can now be reset to multiple start positions

0.3.1

• License updated
• Documentation updated

TODO

• Add 2D environments
• Add documentation

Contributing

Contributions are welcome! Please open an issue or a pull request.

Citation

If you find this project useful, please cite it as:

@misc{Babu_2023,
    author={Babu, Harisankar}, 
    title={Reinforcement Learning Based Path Planning for Autonomous Flight},
    url={https://publica.fraunhofer.de/handle/publica/442672},
    year={2023}, 
    month={Jan},
}

References

• Gymnasium
• OpenAI Gym
• Stable Baselines 3