Maze generation code, inspired by working through Mazes for Programmers. Because the book leans on Object Oriented design (coded in Ruby), much of this is a re-thinking of the approaches in a Clojure style.
Each maze generation algorithm is in its own namespace.
Except where otherwise noted, all algorithms produce "perfect" mazes. Perfect mazes have exactly one path between any two cells in the maze. This also means that you designate any two cells as the start and end and guarantee that there is a solution.
| Usage | Algorithms | Solutions | Utilities | Presentation
Project is pretty much complete and does not have functions exposed for external use (such as through a command-line executable JAR file). All the examples below assume that you are importing into a REPL for execution.
There are several ways to display a maze. The primary data structure used here
to store a maze is a vector of vectors, where each cell indicates which
directions you can navigate out of the cell to. Each of these cells is
position-aware, with cells accessed by [row column].
Here is a 5x5 maze:
(def maze
[[[:east] [:south :west :east] [:west :east] [:west :south] [:south]]
[[:east :south] [:east :north :west] [:south :west] [:north :east] [:west :north]]
[[:north :east] [:west] [:south :north :east] [:west] [:south]]
[[:south] [:south] [:south :north :east] [:west :east] [:west :north :south]]
[[:east :north] [:north :west :east] [:west :north] [:east] [:north :west]]])The easiest way to visualize a maze at the REPL is to generate an ASCII version:
user=> (require '[meiro.ascii :as ascii])
nil
user=> (print (ascii/render maze))
+---+---+---+---+---+
| | |
+---+ +---+ + +
| | |
+ +---+ +---+---+
| | | |
+---+---+ +---+ +
| | | |
+ + + +---+ +
| | |
+---+---+---+---+---+
nilAnd if you want to print or share a maze, it can be output as a PNG:
(require '[meiro.png :as png])
(require '[meiro.core :as m])
(require '[meiro.sidewinder :as sw])
(png/render (sw/create (m/init 15 20)) "sample-maze.png")Which creates a PNG file like:
To print a maze with masked cells:
(def grid (ascii/read-grid "test/meiro/template.txt"))
(require '[meiro.backtracker :as b])
(png/render-masked (b/create grid))
To print a circular (polar) maze:
(require '[meiro.polar :as polar])
(png/render-polar
(b/create (polar/init 10) [0 0] polar/neighbors polar/link))
To print a sigma (hex) maze:
(require '[meiro.hex :as hex])
(png/render-hex
(b/create (m/init 15 20) [7 9] hex/neighbors hex/link))
To print a delta (triangle) maze:
(require '[meiro.triangle :as triangle])
(def grid (ascii/read-grid "test/meiro/triangle.txt"))
(png/render-delta
(b/create grid [0 12] triangle/neighbors m/link))
To print a maze with an inset:
(png/render-inset (b/create (m/init 8 25)) 3)
To print a maze composed of edges, the image must be bored out of a background image. Use the following:
(require '[meiro.prim :as prim])
(def forest (prim/create 25 8))
(png/render-forest forest)
If you want to print an ASCII maze as if it were a series of corridors in NetHack:
(require '[meiro.nethack :as nethack])
(print (nethack/render-corridor maze))
####### # ####### ####### ####### #
# # # # # # # # # # #
### # # ### ### # ### ##### # ### #
# # # # # # # # #
### # ### ### # ### ####### ### ###
# # # # # # # # # #
# # ### # # ### # ### ##### ##### #
# # # # # # # # # #
# # # ### ### ##### # ### ##### # #
# # # # # # # # # # # # #
### ######### ### ####### # #######
If you want to print an ASCII maze as if it were a situated in a NetHack room (corners could use some work):
(print (nethack/render-room maze))
-------------------------------------
|.......|.|.......|.......|.......|.|
|.-----.|.|.-.---.|.-----.|.-----.|.|
|...|.|.|...|...|.|...|.....|.|...|.|
|--.|.|.|------.|.|--.|------.|.---.|
|...|.|...|...|.|...|.......|...|...|
|.---.|--.|.-.|.|--.|------.|.-----.|
|.|.|...|.|.|...|.|...|.....|.....|.|
|.|.|.---.|.|----.|--.|.---------.|.|
|.|.|.|...|...|.....|.|...|.....|.|.|
|.|.|.|.-----.|.---.|.|--.|.-.---.|.|
|...|.........|...|.......|.|.......|
|------------------------------------There are a number of different algorithms for generating mazes.
Binary Tree produces mazes by visiting each cell in a grid and opening a passage either south or east. This causes a bias toward paths which flow down and to the right. They will always have a single corridor along both the southern and eastern edges.
If you wish to generate and print a random binary-tree maze, you can start up a REPL and try to following:
(require '[meiro.core :as m])
(require '[meiro.ascii :as ascii])
(require '[meiro.binary-tree :as bt])
(png/render (bt/create (m/init 8 25)))Which will produce a maze like:
Sidewinder is based upon Binary Tree, but when it navigates south, it chooses a random cell from the current horizontal corridor and generates the link from there. The mazes will still flow vertically, but not to the right as with Binary Tree. All mazes will have a single horizontal corridor along the southern edge.
To generate a maze using the sidewinder algorithm:
(require '[meiro.sidewinder :as sw])
(png/render (sw/create (m/init 8 25)))Which will produce a maze like:
Because Sidewinder creates a maze one row at a time, it is possible to create
infinite mazes. The mazes won't be perfect mazes unless completed, though.
These mazes only link south or east, so you'll only be able to use certain
render functions, like ascii/render and png/render, which are already
optimized to only render the east and south walls per cell. Optional weights can
be passed to the function.
(def infini-maze (sw/create-lazy 25 {:south 2 :east 5}))
(def maze (conj (vec (take 7 infini-maze)) (sw/last-row 25)))
(png/render maze)Which will produce a maze like:
Aldous-Broder picks a random cell in the grid and the moves randomly. If it visits a cell which has not been visited before, it links it to the previous cell. The algorithm ends when all cells have been visited.
Because movement is random, it can take a long time for this algorithm to finish. Because movement is completely random, the generated maze has no bias.
To generate a random-walk maze using Aldous-Broder:
(require '[meiro.aldous-broder :as ab])
(png/render (ab/create (m/init 8 25)))Which will produce a maze like:
Wilson's starts at a random cell and then does a random walk. When it introduces a loop by coming back to a visited cell, it erases the loop then continues the random walk from that point. The algorithm starts slowly, but produces a completely unbiased maze.
To generate a loop-erasing, random-walk maze:
(require '[meiro.wilson :as w])
(png/render (w/create (m/init 8 25)))Which will produce a maze like:
Hunt-and-Kill performs a random walk, but avoids visiting cells which are already linked. When it reaches a dead end but there are still cells to visit, it will look for an unvisited cell neighboring a visited cell and begin walking again from there.
Hunt-and-kill mazes tend to have long, twisty passages with fewer dead ends than most of the algorithms here. It can be slower because it can visit cells many times.
To generate a random-walk maze biased to the first visited cell using Hunt-and-Kill:
(require '[meiro.hunt-and-kill :as hk])
(png/render (hk/create (m/init 8 25)))Which will produce a maze like:
Recursive Backtracker uses a random-walk algorithm. When it encounters a dead end, it backtracks to the last unvisited cell and resumes the random walk from that position. It completes when it backtracks to the starting cell. Resulting mazes have long, twisty passages and fewer dead ends. It should be faster than hunt-and-kill, but has to maintain the stack of all visited cells.
To generate a random-walk maze biased to the last unvisited cell on the path using Recursive Backtracker:
(require '[meiro.backtracker :as b])
(png/render (b/create (m/init 8 25)))Which will produce a maze like:
Kruskal's algorithm is focused on generating a minimum spanning tree. I decided
to use a more graph-centric approach, so the create function returns a
"forest", a map which includes the nodes and edges. It uses x, y coordinates,
so is "backward" from the other algorithms to this point.
The algorithm assigns every cell to a distinct forest, and then merges forests one at a time until there is only one forest remaining.
The png/render-forest function will render a forest directly, or the results
can be converted to the standard, grid-style maze using graph/forest-to-maze
before passing to other png functions.
(require '[meiro.kruskal :as k])
(require '[meiro.graph :as graph])
(def forest (k/create 25 8))
(def maze (graph/forest-to-maze forest))
(png/render maze)Which will produce a maze like:
Prim's algorithm generates a minimum spanning tree by starting with a position
and adding the "cheapest" edge available. Weights are assigned randomly to
ensure a less biased maze. Like Kruskal's, the approach is graph-centric and
create returns a collection of edges. The implementation here is a "True
Prim's" approach, using weighted edges. (There are other versions possible, like
Simplified Prim's, which produce more biased mazes.)
(require '[meiro.prim :as prim])
(require '[meiro.graph :as graph])
(def forest (prim/create 25 8))
(def maze (graph/forest-to-maze forest))
(png/render maze)Which will produce a maze like:
The Growing Tree algorithm is an abstraction over the approach in Prim's
algorithm.
It needs to be passed a queue which holds the active edges of the growing tree
(forest), a poll-fn which removes an edge from the queue, and a shift-fn
which transfers the edges of a newly added node from the set of remaining,
unexplored edges to the queue.
The bias of this algorithm will depend on how edges are added to and removed from the queue
To implement Prim's algorithm using Growing Tree:
(require '[meiro.growing-tree :as grow])
(require '[meiro.prim :as prim])
(def forest (grow/create 25 8
(java.util.PriorityQueue.)
prim/poll
prim/to-active!))
(def maze (graph/forest-to-maze forest))
(png/render maze)Which will produce a maze like:
But, Growing Tree can also be used to implement Recursive Backtracker. Note: If you do not shuffle the new edges, the resulting "maze" will mostly be a series of connected corridors.
(require '[meiro.growing-tree :as grow])
(defn back-poll
[q]
[(first q) (rest q)])
(defn back-shift
[new-edges queue remaining-edges]
(reduce
(fn [[q es] e]
(let [remaining (disj es e)]
(if (= es remaining)
[q es]
[(conj q e)
remaining])))
[queue remaining-edges]
(shuffle new-edges)))
(def forest (grow/create 25 8 '() back-poll back-shift))
(def maze (graph/forest-to-maze forest))
(png/render maze)Which will produce a maze like:
Eller's algorithm processes a row at a time, creating forests as it goes. It also behaves like Sidewinder, in that it will connect to the next row from one random position in a horizontal corridor. When a forest is orphaned, because it does not have a link to the next row, then it is merged with an adjacent forest. When the last row is reached, all forests are merged. Note that when forests are merged, they can be linked at any two adjacent nodes (i.e., not necessarily the southernmost cell).
To create a maze using Eller's:
(require '[meiro.eller :as eller])
(def forest (eller/create 25 8))
(png/render (graph/forest-to-maze forest))Which will produce a maze like:
The Recursive Division algorithm generates fractal mazes and is distinct among all the algorithms here in that is adds walls instead of carving passages.
To create a maze using Recursive Division:
(require '[meiro.division :as division])
(def maze (division/create (m/init 8 25)))
(png/render maze)Which will produce a maze like:
Recursive Division also enables the creation of rooms inside the maze. Do this by passing a maximum room size and a creation rate (a percentage of the time when the subdivision will stop when height and width are below the size).
(def maze (division/create (m/init 8 25) 4 0.4))
(png/render maze)Which will produce a maze like:
To calculate the distance from the north-east cell to each cell using Dijkstra's algorithm:
(require '[meiro.dijkstra :as d])
(def maze (sw/create (m/init 8 8)))
(def dist (d/distances maze))
(print (ascii/render maze (ascii/show-distance dist)))Which will produce a maze like:
+---+---+---+---+---+---+---+---+
| 0 1 | 4 | n | q p | o n |
+ +---+ + +---+ +---+ +
| 1 2 3 | m l | o n | m |
+ +---+---+---+ +---+ + +
| 2 3 | m l | k l | m | l |
+---+ +---+ + +---+ + +
| 5 4 | l k j k | l | k |
+ +---+---+---+ +---+ + +
| 6 | 9 | k j i | h | k j |
+ + +---+---+ + +---+ +
| 7 8 9 a | h g | j | i |
+---+---+ +---+---+ + + +
| g f | a b | g f | i h |
+---+ +---+ +---+ +---+ +
| f e d c d e f g |
+---+---+---+---+---+---+---+---+
To calculate and show a solution:
(def maze (b/create (m/init 8 25)))
(def sol (d/solution maze [0 0] [0 24]))
(print (ascii/render maze (ascii/show-solution sol)))Which will produce a maze like:
+---+---+---+---+---+---+---+---+---+---+---+---+---+---+---+---+---+---+---+---+---+---+---+---+---+
| * | | * * | * * | * * * * * * | * * | * * |
+ + + + +---+ + + + +---+---+---+ + +---+ +---+---+ +---+---+ + + + +
| * | | | * * | * * | * * | | * | | | | * | * * | * | * * | |
+ +---+---+---+ +---+---+---+---+ +---+---+---+ + + + + + + + + +---+---+ +
| * * | * * | * | * * | | * | * * * * | | | | | * * | * * | | |
+---+ + + + + + + + + + +---+---+---+ + + + +---+---+---+---+ + +---+
| * * | * | * | * * | * * | | * | * | | | | | | |
+ +---+ + +---+---+---+ +---+ + +---+---+ + +---+---+ + +---+---+ + +---+ +
| * * * | * | * * * | * * | * | * * | | | | | | | |
+---+---+---+ + + +---+---+ + + + + +---+---+ + +---+---+ + +---+---+ + +
| * * * * | | * * | * | | * | * | * * * | | | | |
+ +---+---+ +---+---+ +---+ + + + +---+---+ +---+---+---+---+ +---+---+ + +---+
| * * * | | * * | * | * * | | * * | * * | * * * * * * * | | | | |
+ +---+ +---+ + + + +---+---+---+---+ + +---+---+---+---+---+---+ + + + + +
| | * * * | * * | * * * * * * | * * * * * * * * | | |
+---+---+---+---+---+---+---+---+---+---+---+---+---+---+---+---+---+---+---+---+---+---+---+---+---+
There are a few additional utilities besides deriving solutions.
Dijkstra's distances calculation can be used to find the position furthest from a given start point. If none is provided, it will assume the upper left-hand corner position.
(d/farthest-pos maze)
[2 20]By running this algorithm twice, the second time with the output of the first run, you can determine the longest path in a maze. This can be useful if you are looking to determine start and end points. This function returns a path with all the positions.
(d/longest-path maze)
([6 20] [6 19] [7 19] [7 20] [7 21] [6 21] [5 21] [5 20] [5 19] [5 18] [6 18]
[7 18] [7 17] [6 17] [6 16] [7 16] [7 15] [6 15] [6 14] [7 14] [7 13] [6 13]
[5 13] [5 14] [5 15] [4 15] [4 16] [3 16] [2 16] [1 16] [1 15] [2 15] [2 14]
[2 13] [3 13] [3 14] [4 14] [4 13] [4 12] [4 11] [5 11] [5 12] [6 12] [7 12]
[7 11] [6 11] [6 10] [7 10] [7 9] [6 9] [6 8] [7 8] [7 7] [7 6] [7 5] [7 4]
[7 3] [7 2] [7 1] [7 0] [6 0] [5 0] [4 0] [3 0] [2 0] [2 1] [1 1] [1 0] [0 0]
[0 1] [0 2] [1 2] [1 3] [1 4] [1 5] [1 6] [1 7] [0 7] [0 8] [1 8] [1 9] [2 9]
[2 8] [2 7] [2 6] [3 6] [4 6] [4 5] [4 4] [5 4] [5 3] [4 3] [3 3] [3 2] [3 1]
[4 1] [5 1] [6 1] [6 2] [6 3] [6 4] [6 5] [5 5] [5 6] [5 7] [4 7] [4 8] [4 9]
[5 9] [5 10] [4 10] [3 10] [3 11] [3 12] [2 12] [1 12] [1 13] [0 13] [0 14]
[0 15][0 16] [0 17] [0 18] [1 18] [2 18] [2 19] [3 19] [3 20] [4 20] [4 21]
[3 21] [2 21] [1 21] [0 21] [0 20] [0 19] [1 19] [1 20] [2 20])By default, the algorithms produce "perfect" mazes, i.e., every position in the grid has one path to any other position in the grid. This inevitably produces dead ends. "Braiding" is the act of removing dead ends from a maze by linking them with neighbors.
To enumerate the dead ends in a maze:
(def maze (b/create (m/init 8 22)))
(m/dead-ends maze)
([0 10] [0 16] [1 1] [1 21] [2 5] [2 13] [3 0] [3 7] [4 2] [4 13] [4 15] [5 3]
[5 10] [6 1] [6 15] [6 19] [7 11] [7 21])You can remove all dead ends with the braid function.
(m/braid maze)
If you don't want to remove all dead ends, you can pass in a rate which will determine what percentage of the dead ends should be removed (randomly).
(def braided (m/braid maze 0.4))
(png/render braided)
Whereas braiding eliminates dead ends by connecting them to neighbors, it is
also possible to cull dead ends, creating a sparse maze. A maze can be culled
multiple times to remove more ends. Culled cells will be marked as masked, so
you will need to use a rendering function which handles this sensibly.
Culled mazes will remain perfect mazes.
(require '[meiro.hunt-and-kill :as hk])
(def maze (hk/create (m/init 8 22)))
(png/render-inset (m/cull (m/cull maze 0.6) 0.6) 3)
A weave maze can connect to non-adjacent cells provided certain conditions are met.
- Passages cannot dead end while underneath another cell.
- Passages must be perpendicular, one north-south, one east-west.
- Passages cannot change direction while traveling under other passages.
A weave maze will need to be rendered using "inset", otherwise it won't be possible to visually identify the under passages.
(require '[meiro.weave :as weave])
(def maze (b/create (m/init 8 25) [0 0] weave/neighbors weave/link))
(png/render-inset maze 2)
Kruskal's is set up to allow weave to be injected into a maze. This is done by pre-seeding the algorithm with cells already combined, and then letting the maze build around it. In order to render a weave maze, it has to be converted to the standard grid format.
(require '[meiro.kruskal :as k])
(require '[meiro.graph :as graph])
(def forests (graph/init-forests 25 8))
(def seeded (reduce k/weave forests
(for [x (range 1 25 2) y (range 1 8 2)] [x y])))
(def forest (k/create 25 8 seeded))
(def maze (graph/forest-to-maze forest))
(png/render-inset maze 2)
The grid-3d namespace can be used to generate three-dimensional mazes.
The example below takes advantage of the controls which can be passed to the
create function to favor spreading out on a level before ascending or
descending.
(require '[meiro.grid-3d :as grid-3d])
(def grid (grid-3d/init 3 4 5))
(def link-3d (m/link-with grid-3d/direction))
(defn select-fn
"Favor selecting neighbors on the same level."
[neighbors]
(let [n (count neighbors)]
(if (and (< 2 n) (< 0.1 (rand)))
(rand-nth (take (- n 2) (rest (sort neighbors))))
(rand-nth neighbors))))
(def maze (b/create grid
(grid-3d/random-pos grid)
grid-3d/neighbors link-3d select-fn))
(png/render-3d maze)
Sometimes you may want to have maze wrap around, meaning if you step off the left edge of the maze, it re-enters on the right edge. You could use this to create a maze on a cylinder. You could use this approach to create a Pac-Man style maze as well.
This will only wrap along the vertical walls:
(require '[meiro.wrap :as wrap])
(def maze (b/create (m/init 8 25) [3 13]
wrap/neighbors-horizontal
wrap/link))
(png/render-inset maze 2)
To wrap off any direction:
(def maze (b/create (m/init 8 25) [3 13] wrap/neighbors wrap/link))
(png/render-inset maze 2)
I did a presentation on this code at Pivotal in January 2018. It is available on YouTube here: Maze Generation Algorithms in Clojure – Michael Daines
Copyright © 2017–2020 Michael S. Daines
Distributed under the Eclipse Public License either version 1.0 or (at your option) any later version.