Implementation:Google deepmind Dm control Variation Colors

Knowledge Sources	Google_deepmind_Dm_control
Domains	Reinforcement_Learning, Domain_Randomization
Last Updated	2026-02-15 04:00 GMT

Overview

Variation classes for randomizing colors in RGB, HSV, and grayscale color spaces, always producing RGBA output arrays for use in MuJoCo visual domain randomization.

Description

The colors module provides three Variation subclasses that allow independent randomization of color channels in different color spaces. All classes produce RGBA numpy arrays as output, making them directly compatible with MuJoCo's rgba attributes on geoms, materials, and other visual elements.

RgbVariation accepts independent variation objects (or constant values) for the R, G, B, and alpha channels. When called, it evaluates each channel variation and assembles them into a four-element RGBA numpy array. HsvVariation operates in the HSV color space, accepting variations for hue, saturation, value, and alpha. It evaluates the HSV components, converts to RGB using Python's colorsys.hsv_to_rgb, and appends the alpha channel. The HSV color space is particularly useful for controlled randomization because hue, saturation, and brightness can be independently varied in perceptually meaningful ways.

GrayVariation is a convenience subclass of HsvVariation that fixes hue and saturation to zero, leaving only the gray level (value channel) and alpha as variable parameters. This simplifies the creation of grayscale color variations.

Usage

Use these classes when building Composer tasks that require visual domain randomization of object colors. HsvVariation is preferred when you want to randomize hue independently from brightness, while RgbVariation gives direct control over individual RGB channels. Bind these variations to MJCF element rgba attributes through a MJCFVariator.

Code Reference

Source Location

Repository: Google_deepmind_Dm_control
File: dm_control/composer/variation/colors.py
Lines: 1-107

Signature

class RgbVariation(base.Variation):
    def __init__(self, r, g, b, alpha=1.0):
        ...

class HsvVariation(base.Variation):
    def __init__(self, h, s, v, alpha=1.0):
        ...

class GrayVariation(HsvVariation):
    def __init__(self, gray_level, alpha=1.0):
        ...

Import

from dm_control.composer.variation import colors
from dm_control.composer.variation.colors import RgbVariation, HsvVariation, GrayVariation

I/O Contract

Inputs (RgbVariation)

Name	Type	Required	Description
r	Variation or float	Yes	Red channel value or variation (0.0 to 1.0)
g	Variation or float	Yes	Green channel value or variation (0.0 to 1.0)
b	Variation or float	Yes	Blue channel value or variation (0.0 to 1.0)
alpha	Variation or float	No	Alpha channel value or variation (default 1.0)

Inputs (HsvVariation)

Name	Type	Required	Description
h	Variation or float	Yes	Hue value or variation (0.0 to 1.0)
s	Variation or float	Yes	Saturation value or variation (0.0 to 1.0)
v	Variation or float	Yes	Value (brightness) value or variation (0.0 to 1.0)
alpha	Variation or float	No	Alpha channel value or variation (default 1.0)

Inputs (GrayVariation)

Name	Type	Required	Description
gray_level	Variation or float	Yes	Gray level value or variation (0.0 to 1.0)
alpha	Variation or float	No	Alpha channel value or variation (default 1.0)

Outputs

Name	Type	Description
return	`numpy.ndarray`	RGBA array of shape (4,) with values in [0.0, 1.0]

Usage Examples

from dm_control.composer.variation import colors
from dm_control.composer.variation import distributions

# Randomize hue while keeping saturation and brightness fixed
random_hue_color = colors.HsvVariation(
    h=distributions.Uniform(0.0, 1.0),
    s=0.8,
    v=0.9,
    alpha=1.0
)

# Randomize each RGB channel independently
random_rgb = colors.RgbVariation(
    r=distributions.Uniform(0.0, 1.0),
    g=distributions.Uniform(0.0, 1.0),
    b=distributions.Uniform(0.0, 1.0)
)

# Create a random gray variation
random_gray = colors.GrayVariation(
    gray_level=distributions.Uniform(0.2, 0.8)
)

# Evaluate the variation
import numpy as np
rng = np.random.RandomState(42)
rgba = random_hue_color(initial_value=None, current_value=None, random_state=rng)
# rgba is a numpy array of shape (4,), e.g. [0.9, 0.18, 0.18, 1.0]

Related Pages

Page Connections

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

Principle

Implementation

Heuristic

Environment