ARCEngine

A Python library for 2D sprite-based game development.

Installation

uv add arcengine
# or
pip install arcengine

Quick Start

ARCBaseGame is the base class for ARCEngine games. Create a game by subclassing it and overriding step():

from arcengine import ARCBaseGame, ActionInput, Camera, GameAction, Level, Sprite

class MyGame(ARCBaseGame):
    def step(self) -> None:
        # Your game logic here.
        # Call complete_action() when you are done handling the input.
        self.complete_action()

level = Level([Sprite([[1]], name="player")])

# Camera is optional (defaults to 64x64).
game = MyGame(game_id="my_game", levels=[level], camera=Camera())

# Multiple frames are returned if an animation was played as a result of the action
frames = game.perform_action(ActionInput(id=GameAction.ACTION1))

API Documentation

ARCBaseGame

Base class for games. Subclass this and implement step().

Properties

• current_level (Level): The current active level
• camera (Camera): The game's camera
• game_id (str): The game's identifier (should be set by subclasses)
• action (ActionInput): The current action being performed
• level_index: int - current level index

Methods

__init__(game_id, levels, camera=None, debug=False, win_score=1, available_actions=[1,2,3,4,5,6], seed=0)

Initialize a new game.

• game_id: Game identifier
• levels: List of levels to initialize the game with (each level is cloned)
• camera: Optional camera to use. If not provided, a default 64x64 camera is created
• debug: Enable debug logging
• available_actions: List of numeric action IDs
• seed: Optional seed value for game logic

Raises ValueError if levels is empty.

debug(message)

Print a debug message if debug mode is enabled.

• message: Message to print

set_level(index)

Set the current level by index.

• index: The index of the level to set as current
• Raises IndexError if index is out of range

set_level_by_name(name)

Set the current level by name.

• name: The level name to match
• Raises ValueError if no level matches

perform_action(action_input, raw=False)

Perform an action and return the resulting frame data.

This method runs step() in a loop until complete_action() is called, rendering each frame. It should not be overridden; implement game logic in step().

• action_input: The action to perform
• raw: If True, returns FrameDataRaw with numpy frames
• Returns: FrameData or FrameDataRaw
• Raises ValueError if an action exceeds 1000 frames

complete_action()

Mark the current action as complete. Call this when the provided action is fully resolved.

is_action_complete()

Check if the current action is complete.

• Returns: True if the action is complete, False otherwise

win()

Call this when the player has beaten the game.

lose()

Call this when the player has lost the game.

handle_reset()

Handle RESET actions, choosing between level_reset() and full_reset() based on action count and ONLY_RESET_LEVELS.

full_reset()

Reset the entire game back to its initial state.

level_reset()

Reset only the current level back to its initial state.

step()

Step the game. This is where your game logic should be implemented.

REQUIRED: Call complete_action() when the action is complete. It does not need to be called every step, but once the action is complete. The engine will keep calling step() and rendering frames until the action is complete.

try_move(sprite_name, dx, dy)

Try to move a sprite and return a list of sprites it collides with.

This method attempts to move the sprite by the given deltas and checks for collisions. If any collisions are detected, the sprite is not moved and the method returns a list of collided sprites.

• sprite_name: The name of the sprite to move
• dx: The change in x position (positive = right, negative = left)
• dy: The change in y position (positive = down, negative = up)
• Returns: A list of sprites collided with. If no collisions occur, the sprite is moved and an empty list is returned
• Raises ValueError if no sprite with the given name is found

Example (try_move):

# Try to move a sprite right by 1 pixel
collisions = game.try_move("player", 1, 0)
if not collisions:
    print("Move successful!")
else:
    print(f"Collided with: {[sprite.name for sprite in collisions]}")

try_move_sprite(sprite, dx, dy)

Try to move a specific sprite and return a list of sprites it collides with.

• sprite: The sprite to move
• dx: The change in x position (positive = right, negative = left)
• dy: The change in y position (positive = down, negative = up)
• Returns: A list of sprites collided with. If no collisions occur, the sprite is moved and an empty list is returned

next_level()

Advance to the next level or mark the game as won if the last level is complete.

on_set_level(level)

Hook called when the level is set. Override to apply level-specific setup.

• level: The level being set

get_pixels_at_sprite(sprite)

Get the camera pixels at a sprite's location.

• sprite: The sprite to sample
• Returns: A numpy array of pixels covering the sprite's area

get_pixels(x, y, width, height)

Get the camera pixels at a given position.

• x, y: Top-left position in camera space
• width, height: Dimensions of the sample area
• Returns: A numpy array of pixels for the given region

Notes

• perform_action runs step() in a loop until complete_action() is called. It raises ValueError if an action exceeds 1000 frames.

API

The public API is exported from arcengine.__init__:

Import necessary components from arcengine:

from arcengine import (
    ARCBaseGame,
    Camera,
    Level,
)

GameAction

Enum of available actions with attached data model types.

• RESET (id 0) uses SimpleAction.
• ACTION1-ACTION5 and ACTION7 use SimpleAction.
• ACTION6 uses ComplexAction to encode screen coordinates x and y (0,0 is the top left pixel). Used for click inputs

Common client/UI conventions:

• ACTION1: Up or W or 1
• ACTION2: Down or S or 2
• ACTION3: Left or A or 3
• ACTION4: Right or D or 4
• ACTION5: Spacebar
• ACTION7: Z - Used for Undo

Sprite

A 2D sprite with position, rotation, scale, and collision behavior.

from arcengine import Sprite, BlockingMode, InteractionMode

# Create a simple 2x2 sprite
sprite_simple = Sprite([
    [1, 2],
    [3, 4]
])

# Create a sprite with custom properties
sprite_custom = Sprite(
    pixels=[[1, 2], [3, 4]],
    name="player",
    x=10,
    y=20,
    layer=1,
    scale=2,
    rotation=90,
    mirror_ud=False,
    mirror_lr=False,
    blocking=BlockingMode.PIXEL_PERFECT,
    interaction=InteractionMode.TANGIBLE,
    # If interaction is None, visible/collidable determine the mode.
    # visible=True, collidable=True are the defaults.
    tags=["player"],
)

Notes

• Pixels are palette indices. Any negative value is treated as transparent when rendering. -1 is treated as transparent and not blocking/non-colliding (applies to BlockingMode.PIXEL_PERFECT) while other negative values are considered blocking.

• Rotation is limited to 0, 90, 180, 270 degrees.

• scale:

  • Positive values upscale (2 = double size, 3 = triple size).
  • Negative values downscale by a divisor: -1 => divide by 2, -2 => divide by 3, etc.
  • 0 is invalid and raises ValueError.

• If interaction is None, visible and collidable determine the interaction mode.

Properties

• name: str
• x: int, y: int
• layer: int
• scale: int
• rotation: int
• blocking: BlockingMode
• pixels: np.ndarray
• interaction: InteractionMode
• tags: list[str]
• mirror_ud: bool, mirror_lr: bool
• is_visible: bool
• is_collidable: bool
• width: int, height: int (based on rendered size)

Methods

__init__(pixels, name=None, x=0, y=0, layer=0, scale=1, rotation=0, mirror_ud=False, mirror_lr=False, blocking=BlockingMode.PIXEL_PERFECT, interaction=None, visible=True, collidable=True, tags=[])

Initialize a new Sprite.

• pixels: 2D list or 2D numpy array representing the sprite's pixels
• name: Optional sprite name (default: generates UUID)
• x: X coordinate in pixels (default: 0)
• y: Y coordinate in pixels (default: 0)
• layer: Z-order layer for rendering (default: 0, higher values render on top)
• scale: Scale factor (default: 1)
• rotation: Rotation in degrees (default: 0)
• mirror_ud, mirror_lr: Optional vertical/horizontal mirroring
• blocking: Collision detection method (default: PIXEL_PERFECT)
• interaction: Optional interaction mode override. If None, visible/collidable determine the mode
• visible, collidable: Used only when interaction is None
• tags: Optional list of string tags

Raises ValueError if scale is 0, pixels is not a 2D list/array, rotation is invalid, or if downscaling factor doesn't evenly divide sprite dimensions.

clone(new_name=None)

Create an independent copy of this sprite.

• new_name: Optional name for the cloned sprite (default: reuses current name)
• Returns: A new Sprite instance with the same properties but independent state

set_position(x, y)

Set the sprite's position.

• x: New X coordinate in pixels
• y: New Y coordinate in pixels

move(dx, dy)

Move the sprite by the given deltas.

• dx: Change in x position (positive = right, negative = left)
• dy: Change in y position (positive = down, negative = up)

set_scale(scale)

Set the sprite's scale factor.

• scale: The new scale factor:
  • Positive values scale up (2 = double size, 3 = triple size)
  • Negative values scale down (-1 = half size, -2 = one-third size, -3 = one-fourth size)
  • Zero is invalid
• Raises ValueError if scale is 0 or if downscaling factor doesn't evenly divide sprite dimensions

For example:

sprite = Sprite([[1, 2], [3, 4]])

# Upscaling examples
sprite.set_scale(2)  # Doubles size in both dimensions
sprite.set_scale(3)  # Triples size in both dimensions

# Downscaling examples
sprite.set_scale(-1)  # Half size (divide dimensions by 2)
sprite.set_scale(-2)  # One-third size (divide dimensions by 3)
sprite.set_scale(-3)  # One-fourth size (divide dimensions by 4)

adjust_scale(delta)

Adjust the sprite's scale by a delta value, moving one step at a time.

The method will adjust the scale by incrementing or decrementing by 1 repeatedly until reaching the target scale. This ensures smooth transitions and validates each step.

Negative scales indicate downscaling factors:

• scale = -1: half size (divide by 2)
• scale = -2: one-third size (divide by 3)
• scale = -3: one-fourth size (divide by 4)

Examples:

• Current scale 1, delta +2 -> Steps through: 1 -> 2 -> 3
• Current scale 1, delta -2 -> Steps through: 1 -> 0 -> -1 (half size)
• Current scale -2, delta +3 -> Steps through: -2 -> -1 -> 0 -> 1

Raises ValueError if any intermediate scale would be 0 or if a downscaling factor doesn't evenly divide sprite dimensions.

set_rotation(rotation)

Set the sprite's rotation to a specific value.

• rotation: The new rotation in degrees (must be 0, 90, 180, or 270)
• Raises ValueError if rotation is not a valid 90-degree increment

rotate(delta)

Rotate the sprite by a given amount.

• delta: The change in rotation in degrees (must result in a valid rotation)
• Raises ValueError if resulting rotation is not a valid 90-degree increment

set_blocking(blocking)

Set the sprite's blocking behavior.

• blocking: The new blocking behavior (BlockingMode enum value)
• Raises ValueError if blocking is not a BlockingMode enum value

set_interaction(interaction)

Set the sprite's interaction mode.

• interaction: The new interaction mode (InteractionMode enum value)
• Raises ValueError if interaction is not an InteractionMode enum value

set_visible(visible)

Set the sprite's visibility.

• visible: The new visibility state

set_collidable(collidable)

Set the sprite's collidable state.

• collidable: The new collidable state

set_layer(layer)

Set the sprite's rendering layer.

• layer: New layer value. Higher values render on top.

set_mirror_ud(mirror_ud)

Set the sprite's mirror up/down state.

• mirror_ud: True to flip vertically

set_mirror_lr(mirror_lr)

Set the sprite's mirror left/right state.

• mirror_lr: True to flip horizontally

set_name(name)

Set the sprite's name.

• name: New name for the sprite
• Raises ValueError if name is empty

render()

Render the sprite with current scale and rotation.

• Returns: A 2D numpy array representing the rendered sprite
• Raises ValueError if downscaling factor doesn't evenly divide the sprite dimensions

collides_with(other, ignoreMode=False)

Check if this sprite collides with another sprite.

The collision check follows these rules:

1. A sprite cannot collide with itself
2. Non-collidable sprites (based on interaction mode) never collide (unless ignoreMode=True)
3. For collidable sprites, the collision detection method is based on their blocking mode:
   • NOT_BLOCKED: Always returns False
   • BOUNDING_BOX: Simple rectangular collision check
   • PIXEL_PERFECT: Precise pixel-level collision detection

• other: The other sprite to check collision with
• ignoreMode: If True, bypasses interaction and blocking checks
• Returns: True if the sprites collide, False otherwise

color_remap(old_color, new_color)

Remap the sprite's color.

• old_color: The old color to remap, or None to remap all colors
• new_color: The new color to remap to

merge(other)

Merge two sprites together.

This method creates a new sprite that combines the pixels of both sprites. When pixels overlap, the non -1 pixels are prioritized

• other: The other sprite to merge with
• Returns: A new Sprite instance containing the merged pixels

BlockingMode

An enumeration defining different collision detection behaviors for sprites:

• NOT_BLOCKED: No collision detection
• BOUNDING_BOX: Collision detection using the sprite's bounding box
• PIXEL_PERFECT: Collision detection using pixel-perfect testing

InteractionMode

An enumeration defining how a sprite interacts with the game world:

• TANGIBLE: Visible and can be collided with
• INTANGIBLE: Visible but cannot be collided with (ghost-like)
• INVISIBLE: Not visible but can be collided with (invisible wall)
• REMOVED: Not visible and cannot be collided with (effectively removed)

Camera

Defines the viewport and renders sprites to a 64x64 output.

from arcengine import Camera

# Create a default camera (64x64 viewport)
camera = Camera()

# Create a custom camera
camera = Camera(
    x=10,                    # X position in pixels
    y=20,                    # Y position in pixels
    width=32,                # Viewport width (max 64)
    height=32,               # Viewport height (max 64)
    background=1,            # Background color index
    letter_box=2,            # Letter box color index
    interfaces=[],           # Optional list of renderable interfaces
)

Notes

• Output is always 64x64. The camera view is uniformly upscaled (nearest neighbor) and letterboxed with letter_box color as needed.
• The scale factor is min(64 // width, 64 // height).
• interfaces is an optional list of RenderableUserDisplay overlays.

Properties

• x: int, y: int
• width: int, height: int (max 64)
• background: int, letter_box: int

Methods

__init__(x=0, y=0, width=64, height=64, background=5, letter_box=5, interfaces=[])

Initialize a new Camera.

Args:

• x (int): X coordinate in pixels (default: 0)
• y (int): Y coordinate in pixels (default: 0)
• width (int): Viewport width in pixels (default: 64, max: 64)
• height (int): Viewport height in pixels (default: 64, max: 64)
• background (int): Background color index (default: 5 - Black)
• letter_box (int): Letter box color index (default: 5 - Black)
• interfaces (list[RenderableUserDisplay]): Optional list of renderable interfaces to initialize with

Raises:

• ValueError: If width or height exceed 64 pixels or are negative

move(dx, dy)

Move the camera by the specified delta.

• dx: Change in x position
• dy: Change in y position

resize(width, height)

Resize the camera viewport.

• width, height: New dimensions (max 64)

render(sprites)

Render the camera view.

The rendered output is always 64x64 pixels. If the camera's viewport is smaller, the view is scaled up uniformly (nearest neighbor) to fit within 64x64, and the remaining space is filled with the letter_box color.

Args:

• sprites (list[Sprite]): List of sprites to render

Returns:

• np.ndarray: The rendered view as a 64x64 numpy array

replace_interface(new_interfaces)

Replace the current interfaces with new ones. This method replaces all current interfaces with the provided ones and stores them as-is (no cloning).

Args:

• new_interfaces (list[RenderableUserDisplay]): List of new interfaces to use. These should be cloned before passing them in.

display_to_grid(display_x, display_y)

Convert display coordinates (64x64) to camera grid coordinates.

• display_x, display_y: Display-space coordinates (0-63)
• Returns: (x, y) grid coordinates, or None if the point lies in the letterbox area

RenderableUserDisplay

The RenderableUserDisplay class is an abstract base class that defines the interface for UI elements that can be rendered by the camera. It is used as the final step in the camera's rendering pipeline to produce the 64x64 output frame.

import numpy as np
from arcengine import RenderableUserDisplay, Sprite

class MyUI(RenderableUserDisplay):
    def render_interface(self, frame: np.ndarray) -> np.ndarray:
        # Modify the frame in-place and return it
        return frame

Methods

render_interface(frame)

Render this UI element onto the given frame.

• frame: The 64x64 numpy array to render onto
• Returns: The modified frame (implementations should modify in-place)

draw_sprite(frame, sprite, start_x, start_y)

Helper to draw a sprite onto a frame with clipping.

• frame: The 64x64 numpy array to draw onto
• sprite: The sprite to draw
• start_x, start_y: Top-left position in frame coordinates
• Returns: The modified frame

ToggleableUserDisplay

The ToggleableUserDisplay class is an example implementation of RenderableUserDisplay that manages a collection of sprite pairs (enabled/disabled states) and provides methods to toggle between them.

from arcengine import ToggleableUserDisplay, Sprite

# Create a toggleable UI element with sprite pairs
ui_element = ToggleableUserDisplay([
    (enabled_sprite1, disabled_sprite1),
    (enabled_sprite2, disabled_sprite2)
])

# Enable/disable specific sprite pairs
ui_element.enable(0)  # Enable first pair
ui_element.disable(1)  # Disable second pair

# Check if a pair is enabled
is_enabled = ui_element.is_enabled(0)

Methods

__init__(sprite_pairs)

Initialize a new ToggleableUserDisplay.

• sprite_pairs: List of (enabled_sprite, disabled_sprite) tuples. Each sprite is cloned.

clone()

Create a deep copy of this UI element.

• Returns: A new ToggleableUserDisplay instance with cloned sprite pairs

is_enabled(index)

Check if a sprite pair is enabled.

• index: Index of the sprite pair to check
• Returns: True if the pair is enabled, False otherwise
• Raises ValueError if index is out of range

enable(index)

Enable a sprite pair.

• index: Index of the sprite pair to enable
• Raises ValueError if index is out of range

disable(index)

Disable a sprite pair.

• index: Index of the sprite pair to disable
• Raises ValueError if index is out of range

enable_all_by_tag(tag)

Enable all sprite pairs that have the given tag.

• tag: Tag to search for

disabled_all_by_tag(tag)

Disable all sprite pairs that have the given tag.

• tag: Tag to search for

enable_first_by_tag(tag)

Enable the first disabled sprite pair with the given tag.

• tag: Tag to search for
• Returns: True if a pair was enabled, False otherwise

disabled_first_by_tag(tag)

Disable the first enabled sprite pair with the given tag.

• tag: Tag to search for
• Returns: True if a pair was disabled, False otherwise

render_interface(frame)

Render the UI element onto the given frame.

• frame: The 64x64 numpy array to render onto

This method renders all sprite pairs, using the enabled sprite if the pair is enabled, and the disabled sprite if the pair is disabled.

Level

Manages a collection of sprites and level metadata.

from arcengine import Level, Sprite, PlaceableArea

sprites = [
    Sprite([[1]], name="player"),
    Sprite([[2]], name="enemy")
]

# Create an empty level
level_empty = Level()

# Create a level with initial sprites
level = Level(
    sprites=sprites,
    grid_size=(16, 16),
    data={"difficulty": "easy"},
    name="level_1",
)

Properties

• name: str
• grid_size: tuple[int, int] | None

Methods

__init__(sprites=None, grid_size=None, data={}, name="Level", placeable_areas=None)

Initialize a new Level.

• sprites: Optional list of sprites to initialize the level with
• grid_size: Optional (width, height) tuple for grid sizing
• data: Optional metadata dictionary
• name: Level name

add_sprite(sprite)

Add a sprite to the level.

• sprite: The sprite to add

remove_sprite(sprite)

Remove a sprite from the level.

• sprite: The sprite to remove

remove_all_sprites()

Remove all sprites from the level.

get_sprites()

Get all sprites in the level.

• Returns: A copy of the sprite list

get_sprites_by_name(name)

Get all sprites with the given name.

• name: The name to search for
• Returns: List of sprites with the given name

get_sprites_by_tag(tag)

Get all sprites that have the given tag.

• tag: The tag to search for
• Returns: List of sprites that have the tag

get_sprites_by_tags(tags)

Get all sprites that have all of the given tags (AND).

• tags: Tags to search for
• Returns: List of sprites with all tags

get_sprites_by_any_tag(tags)

Get all sprites that have any of the specified tags (OR).

• tags: Tags to search for
• Returns: List of sprites that have any tag

get_all_tags()

Get all unique tags from all sprites in the level.

• Returns: A set of tag strings

get_sprite_at(x, y, tag=None, ignore_collidable=False)

Get the top-most sprite at the given coordinates.

• x, y: Coordinates to search
• tag: Optional tag filter
• ignore_collidable: If True, includes non-collidable sprites
• Returns: The first matching sprite or None

collides_with(sprite, ignoreMode=False)

Return all sprites in the level that collide with the given sprite.

• sprite: The sprite to check for collisions
• ignoreMode: If True, bypasses interaction/blocking checks

get_data(key)

Get metadata by key.

• Returns: The stored value or None

clone()

Create a deep copy of this level.

• Returns: A new Level instance with cloned sprites

Development

To set up the development environment:

1. Clone the repository:

   git clone git@github.com:arcprize/ARCEngine.git
   cd ARCEngine

2. Create and activate a virtual environment using uv:

   uv venv
   source .venv/bin/activate  # On Windows: .venv\Scripts\activate

3. Install development dependencies:

   uv sync

4. Install git hooks:

   pre-commit install

This repo uses ruff to lint/format and mypy for type checking:

pre-commit run --all-files

Note: by default these tools run automatically before git commit. It's also recommended to set up ruff inside your IDE (https://docs.astral.sh/ruff/editors/setup/).

Contributions

This project does not accept external contributions.

Citation

If you use this project in your research, please cite it as:

@software{arc_agi,
  author       = {ARC Prize Foundation},
  title        = {ARC Game Engine},
  year         = {2026},
  url          = {https://github.com/arcprize/ARCEngine},
  version      = {0.9.3}
}

License

MIT License

Copyright (c) 2026 ARC Prize Foundation

Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions:

The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.

THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.