Implementation:Farama Foundation Gymnasium Vector Rendering Wrappers

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

Overview

Vector environment rendering wrappers that provide human-mode display and video recording for vectorized environments, merging sub-environment frames into a single composite view.

Description

This module provides two rendering wrappers for vector environments:

HumanRendering:

• Adds "human" render mode support to vector environments that only support "rgb_array" or "depth_array" render modes.
• Merges rendered frames from all sub-environments into a single window using Pygame.
• Automatically computes the optimal grid layout (rows x columns) to fit sub-environments into the screen, maintaining aspect ratio.
• Scales sub-environment renders to fit within the specified screen_size.
• Requires pygame and opencv-python packages.

RecordVideo:

• Records video of vector environment rollouts, concatenating frames from all sub-environments into a single frame.
• Supports episode-triggered and step-triggered recording with configurable schedule functions.
• Finds the optimal frame concatenation layout based on the video_aspect_ratio parameter.
• Supports record_first_only mode to record only the first sub-environment.
• Handles both "next-step" and "same-step" autoreset modes.
• Supports configurable video length, name prefix, FPS, and garbage collection triggers.
• Uses MoviePy for video encoding to MP4 format.
• Requires the moviepy package.

Usage

Use HumanRendering for real-time visualization of vectorized environments during debugging or demonstration. Use RecordVideo for capturing training or evaluation videos for analysis, logging, or sharing.

Code Reference

Source Location

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

Signature

class HumanRendering(VectorWrapper, gym.utils.RecordConstructorArgs):
    def __init__(self, env: VectorEnv, screen_size: tuple[int, int] | None = None): ...

class RecordVideo(gym.vector.VectorWrapper, gym.utils.RecordConstructorArgs):
    def __init__(
        self,
        env: gym.vector.VectorEnv,
        video_folder: str,
        video_aspect_ratio: tuple[int, int] = (1, 1),
        record_first_only: bool = False,
        episode_trigger: Callable[[int], bool] | None = None,
        step_trigger: Callable[[int], bool] | None = None,
        video_length: int = 0,
        name_prefix: str = "rl-video",
        fps: int | None = None,
        disable_logger: bool = True,
        gc_trigger: Callable[[int], bool] | None = lambda episode: True,
    ): ...

Import

from gymnasium.wrappers.vector import HumanRendering, RecordVideo

I/O Contract

Inputs (HumanRendering)

Name	Type	Required	Description
env	VectorEnv	Yes	The vector environment to wrap (must have rgb_array or depth_array render mode)
screen_size	tuple[int, int] or None	No	Screen size for the display window (default: uses sub-env render size)

Inputs (RecordVideo)

Name	Type	Required	Description
env	VectorEnv	Yes	The vector environment to wrap (must have image-based render mode)
video_folder	str	Yes	Directory for saving video recordings
video_aspect_ratio	tuple[int, int]	No	Desired aspect ratio for concatenated frame (default (1, 1))
record_first_only	bool	No	If True, only record the first sub-environment (default False)
episode_trigger	Callable or None	No	Function to determine if recording should start at a given episode
step_trigger	Callable or None	No	Function to determine if recording should start at a given step
video_length	int	No	Max video length in frames; 0 for full episodes (default 0)
name_prefix	str	No	Prefix for video filenames (default "rl-video")
fps	int or None	No	Video frame rate (default: from metadata or 30)
disable_logger	bool	No	Disable MoviePy logging (default True)
gc_trigger	Callable or None	No	When to trigger garbage collection (default: every episode)

Outputs

Name	Type	Description
observation	ObsType	Unchanged observations from the vector environment
rewards	ArrayType	Unchanged rewards from the vector environment
terminations	ArrayType	Unchanged termination flags
truncations	ArrayType	Unchanged truncation flags
info	dict	Unchanged info from the vector environment

Usage Examples

import gymnasium as gym
from gymnasium.wrappers.vector import RecordVideo

# Record video every 5 episodes
envs = gym.make_vec("CartPole-v1", num_envs=5, render_mode="rgb_array")
envs = RecordVideo(
    envs,
    video_folder="save_videos",
    video_aspect_ratio=(1, 1),
    episode_trigger=lambda t: t % 5 == 0,
)
_ = envs.reset(seed=123)
_ = envs.action_space.seed(123)
for i in range(200):
    actions = envs.action_space.sample()
    _ = envs.step(actions)
envs.close()

Related Pages

• Environment:Farama_Foundation_Gymnasium_Python_3_10_Runtime