Visual Fields

sorrel.observation.visual_field.visual_field(world: Gridworld, entity_map: dict[str, ndarray], vision: int | None = None, location: tuple | None = None, fill_entity_kind: str = 'Wall') → ndarray

Visualize the world.

See OneHotObservationSpec.observe() for an example of how this function is used.

Parameters:

  • world – The world tovisualize.

  • entity_map – The mapping between objects and visual appearance.

  • vision – The agent’s visual field radius. If None, the entire environment. Defaults to None.

  • location – The location to center the visual field on. If None, the entire environment. Defaults to None.

  • fill_entity_kind – if the agent’s vision is out of bounds, fill the space with appearances of this entity. Defaults to “Wall”.

Returns:

An array with dtype float64 of shape (number of channels, 2 * vision + 1, 2 * vision + 1). Or if vision or location is None: (number of channels, world.width, world.layers). Here, the number channels is determined based on the one-hot entity map provided.

sorrel.observation.visual_field.visual_field_ascii(world: Gridworld, entity_map: dict[str, str], vision: int | None = None, location: tuple | None = None, fill_entity_kind: str = 'Wall') → ndarray

Visualize the world with ascii appearances.

If the world has multiple layers, and there are multiple non-empty entities on different layers at the same horizontal coordinate, only the top (i.e. highest layer) non-empty entity at that coordinate will be visualized.

See AsciiObservationSpec.observe() for an example of how this function is used.

Parameters:

  • world – The world tovisualize.

  • entity_map – The mapping

  • appearance (between objects and visual)

  • character. (where the visual appearance must be a)

  • vision – The agent’s visual field radius. If None, the entire environment. Defaults to None.

  • location – The location to center the visual field on. If None, the entire environment. Defaults to None.

  • fill_entity_kind – if the agent’s vision is out of bounds, fill the space with appearances of this entity. Defaults to “Wall”.

Returns:

An array of strings of shape (2 * vision + 1, 2 * vision + 1). Or if vision or location is None: (world.height, world.width).