LabMaze is a standalone release of the same maze generator that is used by DeepMind Lab, which is a backtracking maze generator that can be configured to simplify paths and create rooms.

The primary purpose of LabMaze is to provide Python bindings to DeepMind Lab's maze generator, so that the generated maze layout can be used to create navigation tasks through other libraries. This package also includes art assets for texturing the resulting maze environment in the same style as in DeepMind Lab.

The maze generation algorithm and code was originally designed and authored by Charlie Beattie at DeepMind.

If you use LabMaze in your research, we suggest that you cite the DeepMind Lab paper along with a link to the this GitHub repository.

LabMaze contains C++ code that first needs to be compiled into a Python extension. For Python 2.7, 3.5, 3.6, and 3.7 on x64 Linux, macOS, and Windows, we provide built distributions (bdist) that can be straightforwardly installed via:

pip install labmaze

We also upload a source distribution (sdist) to PyPI, which would allow LabMaze to be installed via pip install labmaze on other platforms as well. However, you will first need to install Bazel, along with the requisite platform-specific build toolchain as documented in the linked page.

Alternatively, you can also use pip to install directly from our GitHub repository. You will first need to install Bazel, along with the requisite platform-specific build toolchain as documented in the linked page, then install via

pip install git+git://github.com/deepmind/labmaze

To generate a random maze:

import labmaze
maze = labmaze.RandomMaze(height=11, width=13, random_seed=42)
print(maze.entity_layer)

*************
*   *   *****
*   *   *****
*           *
*** *** *** *
*           *
*   *****   *
*   *****   *
*   *****   *
*   *****   *
*************

By default, maze wall is represented by the '*' token. The RandomMaze object can be re-randomized:

maze.regenerate()
print(maze.entity_layer)

*************
*******     *
*******     *
*     *     *
* *** ***** *
*   *       *
*   *** *** *
*       *   *
*   *   *   *
*   *       *
*************

We can also generate spawn locations and goal object locations, e.g. for use in a navigation task. These locations are also re-randomized along with the maze layout itself when the regenerate() method is called.

maze = labmaze.RandomMaze(height=11, width=13, random_seed=42,
                          spawns_per_room=1, objects_per_room=1)
print(maze.entity_layer)

*************
*P  *P  *****
*   * G *****
* G         *
*** *** *** *
* P      PG *
*   *****   *
*   *****   *
*   *****   *
*G  *****   *
*************

By default, player spawn positions are represented by the 'P' token, and the goal positions are represented by the 'G' token. In the example above, one player spawn position has been generated for each "room" in the maze. One way to use the layout generated above be to pick one of the four spawn points at random, but place an object at all four goal positions.

The RandomMaze object also contains "variation layers" that specifies how different sections of the maze should be textured:

.............
.BBB.DDD.....
.BBB.DDD.....
.BBB.DDD.....
.............
.CCC.....AAA.
.CCC.....AAA.
.CCC.....AAA.
.CCC.....AAA.
.CCC.....AAA.
.............

The '.' token represents the "default" texturing style, while each of the alphabetical token represents distinct styles. It is up to the user to decide how translate the suggested layout into different textures.

LabMaze also supports fixed-layout mazes that are specified via a string:

MAZE_LAYOUT = """
*********
*********
*********
***   ***
***   ***
***   ***
*********
"""[1:]
maze = labmaze.FixedMazeWithRandomGoals(entity_layer=MAZE_LAYOUT)
print(maze.entity_layer)

*********
*********
*********
***   ***
***   ***
***   ***
*********

Spawn and goal positions can also be either specified or randomized into a fixed-layout maze. For example, if one fixed spawn point and one randomized goal is required:

MAZE_LAYOUT = """
*********
***   ***
***   ***
*** P ***
***   ***
***   ***
*********
"""[1:]
maze = labmaze.FixedMazeWithRandomGoals(entity_layer=MAZE_LAYOUT,
                                        num_spawns=1, num_objects=1)
print(maze.entity_layer)

*********
***   ***
***   ***
*** P ***
***   ***
***G  ***
*********

maze.regenerate()
print(maze.entity_layer)

*********
***   ***
***   ***
*** PG***
***   ***
***   ***
*********

More generally, any spawn or goal tokens present in the fixed-layout string are left unmodified by the maze generator. If num_objects is greater than the number of goal tokens already present in the string then additional ones are randomly generated. The same goes for num_spawns.

This is not an official Google product.