Your voice is so soothing that it a-maze-s me

What about cellular automata?

You said each cell is independent, but you still generate one by one?

Now what if you combined all the maze algorithms?

the water and flow analogy for Binary tree maze generation is so beautiful, and the idea to use space filing curves is amazing. Awesome video!

amogus in the thumbnail 💀

latest snapshot has pacman unicode characters in it... perfect for pac-mine

It would be amazing if you could implement origin shift to alter the maze while someone is attempting to complete it- like an actually realized labyrinth. Or, I suppose, you could be extra cruel and just set the maze generator at repeating pulses to generate an entirely new maze multiple times while someone is already in it

If every cell can be computed independently, using a pseudo-random based on the coordinate of the cell and a seed value, the maze could be generated as traversed. Use a height calculated from the coordinate and you can even have a "zero storage" maze, where all that needs to be remembered is your current cell and the random seed. Perfect for an infinite maze.

great video :))

Now I await the 3D tileable maze generator

Your intuition about slopes and paths to use as a basis for maze-generation is very suggestive, and reminds me a lot about conservative fields in vector calculus (vector fields that arise as the gradient of a sloped surface -- basically, they point downwards). I wonder if there's a way to extend this to a much larger class of surfaces, which we can take the gradient of to form basises for new mazes? Taking the gradient of a surface wouldn't be too bad once we had a representation of it, but representing and generating those surfaces in Minecraft would be much harder. You'd have to find a way to encode a loooot of different possible height values. I also don't think every possible random surface would work. Some non-rigorous exploration suggests that valid heightmaps need to have at most one local minima (cell with no neighbors lower than it) -- this is equivalent to the "single origin" condition given in the Origin Shift video. I'm not sure how one could generate such surfaces, without using techniques that are effectively maze-generation methods in and of themselves. For example, you could generate a path, permute it into a non-branching squiggle, then use that as the basis of your maze by turning it into a heightmap... but at that point why not just use origin shift? You could also use some kind of annealing method, where local minima are nudged together until there's only one left. But that would require global information, wouldn't be very fast, and would generally end up with the "origin" at the center. Hmm..

If you run origin shift for long work does it approach uniform spanning tree? Does it actually *reach* uniform spanning tree with the right stopping condition (eg. if you stop the origin has visited each cell you get a uniform spanning tree)?

this is sick! is there a world download available? it'd be really fun to mess around with this

Interestingly, you aren't even limited to using space filling curves for the "heights". The only condition you have is that all cells except for one have at least one neighbor with a lower index (which then probably should be called an elevation rather than an index). Oh, and two neighboring cells have to have different values - but that's often of trivial; for example when even/odd values form a checkerboard pattern. You can for example generate such a height map by taking an existing maze, choosing a random origin point and calculating every cell's distance (along the paths) from that point. The indices of the cells can be pre-calculated from a fixed maze (if you just want to make it slightly less obvious how the maze is generated). Or you can use the maze generated in the previous run. The only downside with that is that the new maze will be quite similar to the previous one. But if you iterate generating a maze from the previous maze an infinite amount of times, you'll probably even get the property that all mazes are equally likely. Fun fact: If you take the taxicab distance from a corner you get back to the binary tree algorithm; And if you take the taxicab distance from the center you get the midpoint tree algorithm.

i dont understand how some insane thing done in minecraft ends up in a video series by a youtuber with less thank 2 k subs! This deserves so much more recognition this is so cool!!!

9:29 erm…

Do you use arrow keys to move or are you able to detect arrow keys in data packs?

Instant repeater?

How about Wilson's, without erasing the loops. It wont make a tree, and sometimes theres gonna be loops (but i think thats interesting for gameplay) but it would be a lot faster

I like recursive division, cus it makes the maze feel like it has "biomes" places where the maze behaves and feels very differently

Origin shift seems like a perfect tool for a Minotaur style shifting maze. Maybe using command blocks to create a giant floating golden cube as the curser of the algorithm (above player height so it doesn’t just crush them).

I hate how great channels like this are not popular but stuff like skibidi tolet has people foaming at the mouth

Very cool

I feel like, if you need to do more painting stuff, you should have all the paintings lined up in your inventory and bring them to your hotbar when it's their turn. Then you would spend less time and mot have to re-check paintings.

All these algorithms generate singly connected mazes that can be solved trivially by following one wall. Multiply connected mazes require the solver to actually memorize where he has been to avoid running in a loop.

I know the title said 1 second maze generator, but that didn't stop my jaw from dropping after seeing it in action.

That was really interesting ❤

This inspired me to start working on a map generator as well, i decided to use spiral tree algorithm, and i so far have gotten a working 8 rt design or 0.8 second generating one. spiral algorithm doesnt generate as difficult ones as hilbert lookahead but the design i have is vertical and each tile is 4x4 with the center always being powered and the 1 layer aorund shared so it works with a standard 2x2 density pixel. im still working on it but so far i have a few pixels working all exactly 0.8 seconds.

Great job! Loved this series!

Can you do a video trying to solve the maze from the inside

What do target blocks do for redstone? I haven't touched modern versions because I play modded, but I see targets everywhere in modern redstone.

I remember seeing the video where CaptainLuma came up with the Origin Shift algo (before the name was even finalized), so seeing people refer to it alongside these much older and more sophisticated algos feels like hearing somebody prove Einstein's general relativity theorem by finding gravitational waves and going "oh yeah, I remember hearing him talk about that in that one Berlin pub in 1916"

One flaw of this: in normal pacman, you can't reverse direction in a hallway, so that you can't just turn away from a ghost to escape, but in this, you can

I think what could be really interesting is creating a tileable cell that combines this with origin shift to allow it to generate the base maze extremely fast but then also make it a changing maze that maintains its solvability. It would also make it more challenging to solve regardless of where the exit is placed

I’ve always made binary division, the downside is the lowest level has dead end that are obvious. The maybe upside is if you don’t see the large view it’s really hard to solve because you can’t tell if you have visited this area before because it locally all looks the same

Binary tree mazes have the property that two people, at the start and finish of the maze, can always meet up in the highest point of Hilbert curve, since it is always possible to walk upwards. This affects the maze quality, but the speed makes up for it.

This is dope

doctor: you have 1 second to live me: thank god I can still generate a random maze with redstone

The main problem I see with this maze is that it still feels really easy to solve. As long as you know there’s a hillbert curve generating the maze you can solve it by trying to follow the curve and trying to “cut corners” as often as possible

The fact you not only managed to make it generate in exactly one second, but you even managed to use the new redstone to your advantage, and on top of that, you made an entire system to replace the configuration using command block just for showcasing the other algorithms... This is mad impressive! A possible feature to add would be to randomize an entrance and exit to the maze (Unless that's already a thing).

That's insanely fast

Very happy to see this and thanks for the shout out!

so to make this in survival I would need to manually select where to place pistons or not for the randomiser. I would have prefered it detecting automatically if there is a valid neighbour

The fact that it's exactly 1s literally makes It Grian-Proof and that is amazong

12:12 that's actually good

Wow, this is insane!

I like how it feels like Wilson’s Algorithm was having a character arc.

Would it be possible to do a different hybrid of aldous-broder-wilson? Rather than having an arbitrary threshold to switch from one to the other, have both methods run at once/alternating steps. The odds of both methods getting stuck at once would be lower, and the transition would be more gradual and convenient. My main concern is if this would still meet the same maze quality result.

Awesome