Algernon-js

Algernon is a JS library for efficiently generating, solving, and rendering 2D mazes.

Where does the name come from?

Found an issue or a performance improvement? Feel free to leave an issue or make a pull request!

Maze generated, solved, and rendered by Algernon-js

• Installation
• Usage
• Benchmarks

Installation

Via NPM:

npm i algernon-js

Via CDN:

<script
	type="module"
	src="https://cdn.jsdelivr.net/npm/algernon-js/dist/algernon.js"
></script>

Usage

Maze Generation

Example:

import { generateBacktrackingRaw } from "algernon-js"

const [rows, cols] = [20, 20]

const rawMaze = generateBacktrackingRaw(rows, cols)

Raw Format:

In algernon-js, mazes are stored as a 2D array of integers for a mix of space efficiency and convenience.

Each cell may have North, South, East, or West walls (the outer edges of the maze will be bounded).

Internally, bits are used to represent the presence of walls or other significant information.

That means a cell is internally used like this (0b1001 - a cell with a North and West wall), but looks like this (9 - the decimal representation).

If you are familiar with bit manipulation, feel free to use mazes in their "raw" format. If not, check out the conversions available for more familiar formats.

Name	Description	Method	Occupancy Grid Support
Backtracking	A fast algorithm for mazes with some long corridors. Reasonable general purpose algorithm.	generateBacktrackingRaw	✅
Kruskal's	On the slower side, good for simple and relatively easy mazes.	generateKruskalRaw	✅
Growing Tree	Fast, performs like Prim's by default but configurable.	generateGrowingTreeRaw	✅

Some of these algorithms are additionally configurable. Check out the JSDoc comments for more info.

Occupancy Grid Format:

Algernon-js also supports the occupancy grid format. In this format, each cell is either open, or a wall/obstacle.

Open is represented as false and a wall/obstacle is true.

There are similar methods available for generating these mazes, like generateBacktrackingGrid.

Note that with this maze format, only odd-dimensioned mazes can be generated, and otherwise the bottom row and right column will both be completely walls. So 19 by 19 would work, but 20 by 20 is not recommended.

Maze Solving

Example:

// maze already generated, stored as `rawMaze`

import { solveAStarRaw } from "algernon-js"

// Solve the maze using A*, starting at the top left
// and ending at the bottom right
const solution = solveAStarRaw(rawMaze, [0, 0], [rows - 1, cols - 1])

Format:

Solutions are simply arrays, with each element as an index in the original rawMaze.

For example: [[0,0],[1,0],[1,1],[1,2],...]

Name	Description	Method	Occupancy Grid Support
A*	Fast Dijkstra's-based solver that uses heuristics. Configurable general purpose solver.	solveAStarRaw	✅
ACO	Ant Colony Optimization is not recommended for real-world purposes. It is interesting for experimentation.	solveACO	⏳
DFS	Depth First Search is simple and fast, exploring deep paths and backtracking.	solveDFSRaw	✅
BFS	Breadth First Search is fast, simply exploring the whole maze.	solveBFSRaw	✅
D* Lite	Essentially A* backwards. Pretty fast, and ideal for mazes that change and require re-planning. See main.js for examples.	solveDStarLite	⏳

Many of these algorithms use heuristics or additional configuration. Check of the JSDoc comments for more info.

Rendering

This is intended as a simple solution for basic web apps or testing. Feel free to use the code as a starting point for a custom solution.

For the occupancy grid format, a similar method is available: renderGridMazeToCanvas.

Example:

<!-- index.html -->

<canvas id="demo-canvas" width="400" height="400"></canvas>

// script.js

import {
	generateBacktrackingRaw,
	solveAStarRaw,
	renderRawMazeToCanvas,
} from "algernon-js"

const [rows, cols] = [20, 20]

// Generate and solve the maze
const rawMaze = generateBacktrackingRaw(rows, cols)
const solution = solveAStarRaw(rawMaze, [0, 0], [rows - 1, cols - 1])

// Render the maze
const canvas = document.getElementById("demo-canvas")
const ctx = canvas.getContext("2d")

// Render each cell at 20 pixels x 20 pixels
// `solution` is an optional argument
renderRawMazeToCanvas(ctx, 20, rawMaze, solution)

Conversion

For the occupancy grid format, ✅ represents support for this format by typically adding Grid to the end of the method (though this may vary if the name implies it).

Raw <-> Node Matrix

// `rawMaze` already generated

const nodeMatrix = convertRawToNodeMatrix(rawMaze)

// A node matrix is a 2D array of `MatrixNode`,
// where each node contains the following booleans:
// hasNorthWall, hasSouthWall, hasEastWall, hasWestWall

// Convert node matrix back to raw
const updatedRaw = convertNodeMatrixToRaw(nodeMatrix)

Raw <-> Node Graph

// `rawMaze` already generated

// Generate a node graph with the given start and end points
const nodeGraph = convertRawToNodeGraph(rawMaze, [0, 0], [row - 1, col - 1])

// A node graph is a graph of `GraphNode`,
// where each node contains the following booleans:
// isStart, isEnd, hasNorthWall, hasSouthWall, hasEastWall, hasWestWall
// and the following references (or null if not present):
// northNeighbor, southNeighbor, eastNeighbor, westNeighbor

const updatedRaw = convertNodeGraphToRaw(nodeGraph, [0, 0])

Raw <-> ArrayBuffer (Binary)

// `rawMaze` already generated

// Convert the maze to a Uint8Array,
// with the first byte storing the rows and cols
const buffer = serializeRawToBinary(rawMaze)

const deserialized = deserializeBinaryToRaw(buffer)

Raw <-> Base64 String

// `rawMaze` already generated

// Convert the maze to a Base64 string
const base64 = serializeRawToString(rawMaze)

const deserialized = deserializeStringToRaw(base64)

Raw -> Supersampled ✅

// `rawMaze` already generated

// Supersample the maze at a factor of 2
const supersampled = supersampleMaze(rawMaze, 2)

// Output: sample maze structure but twice the rows and columns

Raw <-> Grid Maze Format ✅

// `rawMaze` already generated

// Replace wall properties with open/wall cells (and upscale to a factor of 2)
const gridMaze = convertRawToGridFormat(rawMaze, 2)

// Output: same maze structure but walls are represented by true
// and cells by false, and a different number of cells

// Convert back to Raw (and tell it the upscale factor)
const rawMaze2 = convertGridToRawFormat(gridMaze, 2)

// Output: the original `rawMaze`

// Convert points from raw to grid and back
const rawPoint = [2, 2]
const gridPoint = convertRawToGridPoint(rawPoint)
const originalPoint = convertGridToRawPoint(gridPoint)

Raw -> Raw (Braided) ✅

// `rawMaze` already generated

// Remove dead-ends with the given probability
braidMaze(rawMaze, 0.5)

// Output: an in-place modification of the original maze,
// with walls removed to prevent dead-ends.

Grid -> Degrade/Fill ✅

// `gridMaze` already generated

// Remove walls with the given probability
degradeGrid(gridMaze, 0.1)

// Add walls with the given probability
fillGrid(gridMaze, 0.05)

// Output: an in-place modification of the original maze,
// with walls removed or added. This is ONLY for occupancy
// grid mazes.

Helpers

Internally used helper methods, like removeWall, getAvailableNeighbors, getDirection, and more, are made available through the helpers namespace.

Additionally, many have support for occupancy grids as well by adding Grid to the end of the method name.

Example:

import { helpers } from "algernon-js"

const c1 = [0, 0]
const c2 = [1, 0]

const directionBetweenCells = helpers.getDirection(c1, c2)
// directionBetweenCells: 0b0100 or South

Data Structures

Internally used data structures are now available as well.

The two currently available are: createDisjointSet and createMinHeap.

Benchmarks

Benchmarks were run on a M2 MacBook Air with a minimum of 50 samples each.

Generation

Maze Size	Name	Mean (ms)	RME
20 x 20	Backtracking	0.1037	0.89
50 x 50	Backtracking	0.5939	0.80
80 x 80	Backtracking	1.4800	0.70
20 x 20	Kruskal	2.0065	1.49
50 x 50	Kruskal	44.493	3.59
80 x 80	Kruskal	325.52	2.84
20 x 20	Growing Tree	0.1283	0.39
50 x 50	Growing Tree	0.8033	0.70
80 x 80	Growing Tree	2.1503	2.45

Solving

Backtracking mazes were selected for this benchmark. Performance can vary with different maze types.

Maze Size	Name	Mean (ms)	RME
20 x 20	A-Star	0.0173	1.13
50 x 50	A-Star	0.1551	0.81
80 x 80	A-Star	0.1676	0.78
20 x 20	ACO	1.3462	0.64
50 x 50	ACO	22.810	0.98
80 x 80	ACO	69.850	1.67
20 x 20	DFS	0.0617	0.41
50 x 50	DFS	0.2197	0.48
80 x 80	DFS	0.3464	0.60
20 x 20	BFS	0.0532	0.38
50 x 50	BFS	0.1643	0.62
80 x 80	BFS	0.1504	0.37
20 x 20	D* Lite	0.1215	1.06
50 x 50	D* Lite	0.5553	0.64
80 x 80	D* Lite	0.8571	1.71