Implementation:Farama Foundation Gymnasium Vector ArrayConversion

Knowledge Sources	Farama_Foundation_Gymnasium Gymnasium Docs
Domains	Reinforcement_Learning, Wrappers
Last Updated	2026-02-15 03:00 GMT

Overview

A vector environment wrapper for converting between Array API compatible frameworks (numpy, torch, jax.numpy, cupy, etc.), serving as the vectorized counterpart of the single-environment ArrayConversion wrapper.

Description

The ArrayConversion vector wrapper converts all inputs and outputs of a vector environment between two Array API compatible frameworks. Unlike the single-environment version, this wrapper also converts rewards, terminations, and truncations (which are arrays in vector environments) rather than casting them to Python scalars.

The wrapper:

• Converts actions from the target framework to the environment framework before stepping.
• Converts observations, rewards, terminations, truncations, and info from the environment framework to the target framework after stepping and resetting.
• Converts options to the environment framework during reset.
• Supports pickling via __getstate__/__setstate__ with string-based module name serialization.

Uses the array_conversion singledispatch function from gymnasium.wrappers.array_conversion for the actual conversions.

Usage

Use this wrapper when you have a vectorized environment using one Array API framework but your training code uses another. This is the vector equivalent of gymnasium.wrappers.ArrayConversion.

Code Reference

Source Location

• Repository: Farama_Foundation_Gymnasium
• File: gymnasium/wrappers/vector/array_conversion.py

Signature

class ArrayConversion(VectorWrapper, gym.utils.RecordConstructorArgs):
    def __init__(
        self,
        env: VectorEnv,
        env_xp: ModuleType | str,
        target_xp: ModuleType | str,
        env_device: Device | None = None,
        target_device: Device | None = None,
    ): ...

Import

from gymnasium.wrappers.vector import ArrayConversion

I/O Contract

Inputs

Name	Type	Required	Description
env	VectorEnv	Yes	The Array API compatible vector environment to wrap
env_xp	ModuleType or str	Yes	The Array API framework the environment uses
target_xp	ModuleType or str	Yes	The Array API framework to convert outputs to
env_device	Device or None	No	The device the environment is on (default None)
target_device	Device or None	No	The device for output arrays (default None)

Outputs

Name	Type	Description
observations	target_xp array	Observations converted to target framework
rewards	target_xp array	Rewards converted to target framework array
terminations	target_xp array	Termination flags converted to target framework array
truncations	target_xp array	Truncation flags converted to target framework array
info	dict	Info dict with values converted to target framework

Usage Examples

import numpy as np
import gymnasium as gym
from gymnasium.wrappers.vector import ArrayConversion

# Convert a JAX vector environment to return NumPy arrays
envs = gym.make_vec("JaxEnv-vx", num_envs=3)
envs = ArrayConversion(envs, env_xp="jax.numpy", target_xp="numpy")
obs, info = envs.reset(seed=123)
type(obs)  # <class 'numpy.ndarray'>

Related Pages

• Environment:Farama_Foundation_Gymnasium_Python_3_10_Runtime