unienv 0.0.1b10

Unified robot environment framework supporting multiple tensor and simulation backends

UniEnv

Framework unifying robot environments and data APIs. UniEnv provides an universal interface for robot actors, sensors, environments, and data.

Tensor library cross-backend Support

UniEnv supports multiple tensor backends with zero-copy translation layers through the DLPack protocol, and allows you to use the same abstract compute backend interface to write custom data transformation layers, environment wrappers and other utilities. This is powered by the XBArray package.

Universal Robot Environment Interface

UniEnv supports diverse simulation environments and real robots, built on top of the abstract environment / world interface. This allows you to reuse code across different sim and real robots.

Universal Robot Data Interface

UniEnv provides a universal data interface for accessing robot data through the abstract BatchBase interface. We also provide a utility ReplayBuffer for saving data from various environments with diverse data format support, including hdf5, memory-mapped torch tensors, and others.

Installation

Install the package with pip

pip install unienv

You can install optional dependencies such as gymnasium (for Gymnasium-compatible environments), dev, or video by running

pip install unienv[gymnasium,video]

Local Developments

Development Environment Setup

To perform development on your local machine, you need to clone the repository and install the package in editable mode.

git clone https://github.com/UniEnvOrg/UniEnv
cd UniEnv
pip install numpy
pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cpu # You can choose to either install cpu version or cuda version, up to you
pip install jax # same for jax
python -m pip install pytest
python -m pip install tensordict h5py opencv-python
pip install -e .[dev,gymnasium,video]

Before commiting

Make sure all unit tests pass and your added code compiles before commiting or making a PR. You can run the tests with

pytest

Cite

If you use UniEnv in your research, please cite it as follows:

@software{cao_unienv,
  author = {Cao, Yunhao AND Fang, Kuan},
  title = {{UniEnv: Unifying Robot Environments and Data APIs}},
  year = {2025},
  month = oct,
  url = {https://github.com/UniEnvOrg/UniEnv},
  license = {MIT}
}

Acknowledgements

The idea of this project is inspired by Gymnasium and its predecessor OpenAI Gym. This library is impossible without the great work of DataAPIs Consortium and their work on the Array API Standard. The zero-copy translation layers are powered by the DLPack project.