Implementation:Farama Foundation Gymnasium SyncVectorEnv

Knowledge Sources	Gymnasium SyncVectorEnv
Domains	Reinforcement_Learning, Parallelism
Last Updated	2026-02-15 03:00 GMT

Overview

Concrete tool for running multiple environments sequentially in a single process provided by the Gymnasium library.

Description

SyncVectorEnv creates and manages multiple environment instances in the main process. It instantiates environments from callables, batches their spaces using batch_space, and provides a unified VectorEnv interface. The copy parameter controls whether observations are copied (True, safe) or returned as views (False, faster but aliased).

Usage

Use SyncVectorEnv directly when you need custom environment factories with different configurations per sub-environment, or via gymnasium.make_vec with vectorization_mode="sync".

Code Reference

Source Location

Repository: Gymnasium
File: gymnasium/vector/sync_vector_env.py
Lines: L26-120

Signature

class SyncVectorEnv(VectorEnv):
    def __init__(
        self,
        env_fns: Iterator[Callable[[], Env]] | Sequence[Callable[[], Env]],
        copy: bool = True,
        observation_mode: str | Space = "same",
        autoreset_mode: str | AutoresetMode = AutoresetMode.NEXT_STEP,
    ):
        """Vectorized environment that serially runs multiple environments.

        Args:
            env_fns: Iterable of callable functions that create the environments.
            copy: If True, reset and step return copies of observations.
            observation_mode: "same" (identical spaces) or "different" (varying bounds).
            autoreset_mode: How to handle automatic resets on episode end.
        """

Import

from gymnasium.vector import SyncVectorEnv
import gymnasium as gym

envs = SyncVectorEnv([
    lambda: gym.make("CartPole-v1"),
    lambda: gym.make("CartPole-v1"),
])

I/O Contract

Inputs

Name	Type	Required	Description
env_fns	Sequence[Callable]	Yes	Factory functions that create environments
copy	bool	No	Copy observations (default True)
observation_mode	str	No	"same" or "different" (default "same")
autoreset_mode	AutoresetMode	No	Autoreset behavior (default NEXT_STEP)

Outputs

Name	Type	Description
num_envs	int	Number of sub-environments
single_observation_space	Space	Observation space of one sub-environment
single_action_space	Space	Action space of one sub-environment
observation_space	Space	Batched observation space
action_space	Space	Batched action space

Usage Examples

Heterogeneous Environments

from gymnasium.vector import SyncVectorEnv
import gymnasium as gym

# Different gravity settings per environment
envs = SyncVectorEnv([
    lambda: gym.make("Pendulum-v1", g=9.81),
    lambda: gym.make("Pendulum-v1", g=1.62),  # Moon gravity
    lambda: gym.make("Pendulum-v1", g=3.72),  # Mars gravity
])

obs, infos = envs.reset(seed=42)
print(obs.shape)  # (3, 3) - 3 envs, 3-dim observation
envs.close()

Related Pages

Page Connections

Double-click a node to navigate. Hold to expand connections.

Principle

Implementation

Heuristic

Environment