Sunflower Maze Maker

What did I learn

After visiting a sunflower maze with my partner and little boy, I became curious about how these mazes are planned. This curiosity led me to create my own maze generator. Interestingly, while on the ground the maze seemed really intricate... but an aerial view told a different story!

Here is the image of the sunflower field we went too:

Admittedly, I only received this image from my partner after I was already building my maze generator - but the challenge had already been set.

So the first thing I learnt is that there are many different types of maze algorithms. It is more common for a maze generator to create a single path solution.

One of the first things I learned was that there are many different maze-generating algorithms. Most generate a single solution path, which isn’t ideal for real-world mazes like sunflower or hedge mazes. In these, if the first few visitors find the exit, those behind them simply follow. This turns a maze into a conga line of followers.

Thats why a maze with multiple players should ideally have many paths from start to finish.

To encourage people to go down all paths, a minigame is usually set up where you have to find all the QR codes throughout the maze, and if you scan all of them you'll win something or you'll be entered into a draw etc.

With this in mind, I created a playful and chaotic solution: the "drunk worm algorithm." Here's how it works:

The algorithm starts with a defined space. A 'worm' begins at a random unvisited node and moves in random directions, carving a path as it goes (inspired by The Drunkard's Walk). If a worm hits a wall or itself, it stops, and a new worm starts from another random point.

Once all nodes are visited or surrounded by already-visited spots, the process stops. Any isolated pillars are connected to their nearest neighbor to prevent dead zones.

This leaves a problem. We've just kind of randomised everything! There are walls being built and broken down so randomly, that its possible that the maze might not be solvable!

The next step then, does a recursive search through to see what nodes can't be visited. If a node can't be visited, a wall shielding off those nodes is removed, and the recursive function continues from where that wall was removed.

Finally we have a maze made by our drunk worms which is solveable.

After running this algorithm several times (try clicking refresh), you’ll see a variety of maze patterns—some interesting, others a little chaotic.

To visualize how people might traverse the maze, I also created a 'walker' simulation. Initially, these walkers moved randomly but took far too long to find the exit. So, I gave them a memory, allowing them to prioritize unexplored areas while navigating the maze. By releasing multiple walkers, I could simulate how a crowd might move through the maze.

What is next for this project

I would like to have a different groups of tesselatable objects availablle such as triangles, squares, hexagons etc. However this can be a rabbit hole to go down as you can have different shapes which put together create a tesselatable group (such as an octagon + square. But you can also have a hexagon + pentagon but this causes folding which makes a football retain its shape.

I would like to have inputs for people to enter in their own spaces, and configure it so the client can determine how thick they want the wall to be (too thin and people just smash through it).

I'd like to put in some analysis of the walkers. How long it took the walker, heat map to show where they went on the map. Standard deviation across the walkers.

I'd like to use the lessons learnt from the Github World project to make a 3D projection of what the walkers can see. And then allow you to click through the maze