Brian Bucklew - Dungeon Generation via Wave Function Collapse

This talk is from the 2019 Roguelike Celebration - find out more at roguelike.club/event2019.html
Brian is a developer at Freehold Games and has worked on Caves of Qud and Sproggiwood

[@antindie]

Audio - the unsolvable problem.

[@Achelon]

Put the speakers in front of the stage directed towards the audience. Use a proper microphone with lets say a heart shaped input. Then a person who understands what is gain and volume. If these 4 things are done it is extremely easy to get good audio anywhere. When some of these are lacking, the person has to be extremely close to the microphone, or the microphone/speaker stand has to be situated into a place where the speakers cannot blow right into it. I worked at a place where this was a problem when the stage was lifted up, it was super annoying because training people who are speaking was not really possible.

[@nicksouthcote8826]

@@Achelon did you mean cardioid?

[@arcan762]

For anyone wanting a better quality version of this exact same talk (same presentation with same slides etc.) search for "Tile-Based Map Generation using Wave Function Collapse in 'Caves of Qud'"

[@nickmaxwellambient6615]

Brian Bucklew is the man. One of my favorite developers.

[@nexovec]

The audio sounds like if it was procedurally generated :D

[@Osse22]

Amazing lecture, thanks for posting.

[@juantesino7764]

Live and Drink!

[@westwind1059]

I knew it! I've got a hunch, reading about how there were some Golgotha levels below final one and it looked to me like a function with smooth transgretion of values.

[@cromdesign1]

I tried before to create my own version but failed. I watched this video and found a tutorial online and was able to create a working version. WFC in its base seems like a simple auto tiling algorithm. The descriptions of how it works in all its forms are quite complex.

[@f0kes32]

Did you do an overlapping version?

[@f0kes32]

Can you give a link to the tutorial?

[@pedroprobst5230]

It's indeed a hard algorithm to implement from scratch.

[@benjaminpattison1907]

@@pedroprobst5230 I can attest. I've been trying to port the original C# to GDScript for the Godot engine.

[@MrAthoOome]

wow that's insane

[@RobertoMaurizzi]

Tile constraining would be much better to generate "finished" and more human-like maps, since you can put doors, windows, furniture... In addition the output size and so algorithm memory and time use tends to be smaller (lower number of bigger tiles)

[@heroclashes941]

can somebody provide link or implementation of it because I don't think I've heard of such thing

[@tullman3craggs102]

Damn I got caves of cud it's good game awesome to hear from the Dev seems knowledgeable

[@Hemebond]

I spy a Dungeonmans

[@djbanizza]

fantastic algorithm

[@42yeah]

He's got magnificent beards

[@ItsJustAstronomical]

This algorithm was not invented by Maxim Gumin. It is the same as the Model Synthesis algorithm published almost a decade earlier by Paul Merrell.

[@dandymcgee]

Is that the Github office?? I swear I've been in this room before. Haha.

[@TheBlenderblob]

how is 'wave function collapse' different to an n dimensional markov chain?? litteraly name one difference please.

[@jouethrougha7542]

none of these words are in the bible

[@diadetediotedio6918]

Read the paper bro. But in short, to name one difference, markov chains are memory-less and based on matching neighbors sequentially (generally). This algorithm is about entropy reduction via collapsing cells, in terms of a markov chain you can think of it as having the matching pairs but instead of just sampling them, you sample the cells with the more stable information one-by-one until the entire map collapses.

[@voidling2632]

:( can't understand a word

[@cr6925]

Looks interesting. If only I could hear it. Gave up. Did you not think to check the audio before posting?

[@andrewherrera7735]

I thought machine learning in unity would be used for stuff like this but all I see is people using it for lame things like pong or driving a car on a road.

[@MrSharmabros]

If you knew more about ml you wouldn't think so