RGBBox

class RGBBox(shape=(3,), *args, init_value=None, **kwargs)[source]

Bases: Box

Specialized box space for RGB image data with automatic constraints.

This class extends Box to provide a convenient space for RGB images, automatically enforcing uint8 dtype and [0, 255] value ranges. It validates that the shape includes exactly 3 channels for RGB data.

Parameters:

  • shape (tuple) – Shape of the image. Must include a dimension of size 3 for the RGB channels. Common formats: (H, W, 3) or (3, H, W).

  • init_value (np.ndarray, optional) – Initial RGB image. Must match shape and be uint8 dtype.

  • *args – Additional positional arguments passed to Box.

  • **kwargs – Additional keyword arguments passed to Box.

Variables:

  • init_value (np.ndarray) – The initial RGB image.

  • value (np.ndarray) – The current RGB image.

Example

Create a space for 64x64 RGB images:

import numpy as np

space = RGBBox(
    shape=(64, 64, 3), init_value=np.zeros((64, 64, 3), dtype=np.uint8)
)
image = space.sample()  # Random RGB image
space.reset()  # Returns to black image
Raises:

AssertionError – If shape does not contain a dimension of size 3.

Note

This space is useful for vision-based environments where images need to be sampled or tracked as part of environment configuration. The low, high, and dtype parameters are automatically set and cannot be overridden.