Getting Started

This page is the fastest way to orient yourself in the UniEnvPy package.

Install

Install the base package:

pip install unienv

Install optional extras when you need them:

pip install "unienv[gymnasium,video]"

The project also exposes optional integrations and storage backends that depend on external packages such as torch, jax, pyarrow, h5py, datasets, or huggingface_hub. Install those separately when you use the matching modules.

The Two Main Namespaces

• unienv_interface: online interaction APIs such as spaces, environments, worlds, wrappers, and transformations
• unienv_data: offline data APIs such as batches, replay buffers, storage backends, samplers, and dataset adapters

If you are only building environments, most of your work will start in unienv_interface.

If you are only building dataset or replay-buffer tooling, most of your work will start in unienv_data.

Mental Model

UniEnv is easiest to understand as four layers:

1. Spaces describe what valid data looks like.
2. Environments and worlds produce or consume that data over time.
3. Wrappers and transformations reshape the interface without changing your core implementation.
4. Replay buffers and dataset adapters store, sample, and serve the resulting trajectories or transitions.

Smallest Useful Example

import numpy as np

from unienv_interface.backends.numpy import NumpyComputeBackend
from unienv_interface.space.spaces import BoxSpace, DictSpace
from unienv_data.replay_buffer import ReplayBuffer
from unienv_data.storages.parquet import ParquetStorage

backend = NumpyComputeBackend

transition_space = DictSpace(
    backend,
    {
        "obs": BoxSpace(backend, 0.0, 1.0, np.float32, shape=(4,)),
        "action": BoxSpace(backend, -1.0, 1.0, np.float32, shape=(2,)),
    },
)

buffer = ReplayBuffer.create(
    ParquetStorage,
    transition_space,
    cache_path="cache/demo_buffer",
    capacity=1_000,
)

print(buffer.single_space)

This already shows the basic pattern:

• define the structure of one sample with a Space
• build storage around that space
• keep higher-level code generic over the exact backend or storage implementation

Where To Go Next

• Read Environments if you are implementing runtime interaction.
• Read Spaces and Backends if you need structured data validation or cross-backend movement.
• Read Replay Buffers and Storage if your next step is logging or offline learning.

Local Development

Inside the repository:

pip install -e .[dev,gymnasium,video]
pytest

Documentation Build

This site is built from hand-written Markdown in docs/ plus generated API pages for unienv_interface and unienv_data.

To serve it locally after installing the MkDocs toolchain:

mkdocs serve