YT YouTube2Text

YouTube2Text → Transcripts → EPC2018 | Wave Function Collapse in Bad North | Oskar Stalberg

EPC2018 | Wave Function Collapse in Bad North | Oskar Stalberg — Transcript

by BUas Games · 6,335 words · 880 segments · language en · Watch on YouTube

Open in transcriber Summarize with AI Download SRT/VTT

Full transcript

0:00all right hello everyone nice to be here
0:03yeah my name is Oscar solver I make
0:07procedural planets and pretty things I
0:10come from like an artist's background I
0:13studied at the game assembly mama which
0:15is a single place similar to this one I
0:17studied art there but then I should have
0:19ventured more and more into technical
0:21science and making more procedural
0:23things I used to work at massive I
0:26worked on the division I made that cool
0:29like holographic map maybe you've seen
0:30it I was good fun and side projects I've
0:36got a bunch of other like procedural
0:38interactive demos just to explore things
0:41myself but and I mean they habit of
0:42putting them out this interactive demos
0:44online too so both that little house
0:46there called brick block and not planet
0:49you can find on my web page and you can
0:50open them in the browser and it's just
0:52like tools to manipulate a house and
0:54manipulate a planet that's a very good
0:59way to do projects as student too by the
1:01way
1:01just like make things that you can
1:03publish that you don't just publish
1:04screenshots or videos that publish like
1:06interactive things because there's so
1:08much more likely to be shared and it's
1:10so much more impressive so do that
1:13yeah I said I worked with NASA for a few
1:16years I worked at us to model and say
1:19for a bit too but now I'm indeed making
1:23my own games well trying to make my own
1:26game one game called bad north looks
1:29like this right now but today I am going
1:33to talk to you about a procedural
1:38algorithm that we are using to generate
1:40these levels it's called weight function
1:42collapse I didn't name it wave function
1:44collapse so you know don't ask me about
1:47the connection to quantum mechanics but
1:49this is like a gift I came across this
1:51guy Maxim goomy is the guy who invented
1:54the concept or I mean he based it off of
1:57something else too but I think he did
1:59the bulk of it you mentioned so as you
2:02can see it's well the way he did it is a
2:05way to use like a small example bitmap
2:08and they create a larger bitmap that's
2:11consistent with that so what it really
2:13does it's like deriving adjacency
2:16constraints like which pixels are
2:17allowed to be next to which pixels from
2:20the example and then it can create
2:21something bigger out of that so
2:23obviously my first idea when I saw this
2:25one was like I bet you can use this for
2:273d I bet you can use this for other
2:28things than just textures to build
2:33little larger works because it tied him
2:35very neatly to like some problems I had
2:37looked at before
2:38like how to make nicely very sort of
2:41Compton's agnostic procedural algorithms
2:47so yeah this is the first little thing
2:53all right yeah there we go yeah he's had
2:58to download these is like half of these
3:00are just my old tweets that I downloaded
3:02again into the you should follow me on
3:05Twitter to a post about these kind of
3:06things but this is like my first
3:08implementation I hope the wave function
3:10cooks that I built and as you see it
3:12what it does it it's applied on like a
3:15triangular tiles model on spheres one in
3:19the middle there is like villages roads
3:24quite similar to the planet I did before
3:26but it can also it can do some things
3:28that that planet couldn't do like some
3:31ways of making roads and because that
3:32pipe didn't have worlds for examples
3:34they like are a bit different
3:35otherwise but you can also make you do
3:37like weird things you see this guy's
3:39just like building sort of also just to
3:43yeah I don't like UT I don't want the
3:45triangular tiles for that would look
3:46like this video doesn't seem too long
3:49but that's just another night concert so
3:51the core idea is that you just you just
3:53give it a title set and then it figures
3:56out how to assemble that tile set into
3:59one coherent model or environments
4:02people like whatever constraints are
4:04implied by how your tile set fits
4:07together it will solve for that so if
4:10you give it like if you if you if you
4:14haven't build a ton of transitions it
4:16will just figure out a way to build the
4:19environment with those transitions that
4:21you have did you have that build so this
4:24is another early example of what I did
4:26there so you see it's like building a
4:27little sort of safe walk something like
4:28that and these like to the left third
4:31that's actually all the inputs modules
4:33been used so it's not today it's quite
4:35quick to get started with it and then
4:37you figures out how to put it together
4:39into different things and this I think
4:45this is the album is algorithm is my
4:47contents agnostic fundamentally like it
4:49doesn't know that it's building a house
4:50it just knows that it needs to fit
4:52things together it makes it like a Civ
4:55artist you can build all these nice the
4:57whole sort of transitions between plates
4:59and you can build like extra little
5:01things use things like you can build a
5:03chimney and you can build like I have
5:06some larger peach some sort of extra
5:08features of my been going up here these
5:12little sort of divider here so yeah you
5:15can build things like that and then it
5:17will like three larger structures form
5:18it so now I'm going and this is another
5:21thing I tried to build this is actually
5:23invisible became I had worth a long time
5:25ago it looks like this and you played a
5:26little cat jumped around but then but
5:29that all changed it turned up then this
5:31is an important lesson it was too
5:32colorful it's very hard to put gameplay
5:35on top of an environment that's very
5:36colorful because then the gameplay is
5:38even more colorful
5:39unlike the UI
5:41the attention but it's also just an
5:44example of the kind of variation you can
5:45get from from the algorithm so now we're
5:50going to get into sort of the specifics
5:52of the algorithm how it actually works
5:55and as any of you looked at this other
5:58working before few people have that's
6:02good so my hope is here that actually
6:04has any of you built one yourself for
6:06trying to a few people that's really
6:10cool so my hope is that actually like if
6:11you've been interested in it you should
6:14from this presentation be able to get
6:16going and start sort of understanding
6:19the core ok so this is a very simplified
6:21version of it that's made specifically
6:24for presentations so you see here
6:27there's like the input modules and here
6:30is like a big possibility space and call
6:32cuz there's a bit and we do it too when
6:33you talk about tiles like it's a title
6:35where things go or is a tile what you
6:38place there so I split up the
6:39terminology into modules and slots so
6:42like come on each slot picks a module
6:44that's the terminology so right now you
6:47have everything being possible every
6:50single module here being possible inside
6:52each slot and I can just quick solve
6:55what so you see it like constricts
6:59different environments there our
7:02therapist is like how the particles
7:03input modules and this like this
7:05particular set of module because the
7:07algorithm can fail in some cases but
7:09with this input this input module it
7:12can't actually sell because it's too
7:13simple there are no what we're
7:14dependence is built into it and the way
7:18it works
7:18let's start by like these selecting so
7:23now I will only feed it to modules only
7:26too much was possible full sky or full
7:28dark so obviously since they don't have
7:30any transition between those modules I
7:32don't have any taught us that I'd like
7:33both sky on both dirt as soon as I
7:35select one of them everything turns into
7:38sky for example or everything turns into
7:41dirt because there's nothing like
7:43there's nothing to connect
7:44but if I had like this little connector
7:47here right so it's dirt at the bottom
7:49this kind of top if I slip sky here yeah
7:52everything above it I'm inviting that it
7:54has to be sky
7:55if I select dirt here yeah everything
7:57along with that and below that has to be
7:59dirt but here from still a bit ambiguous
8:00but as soon as I place the like the top
8:05line part there then like the entire
8:07class is done so the way and of course
8:15yeah you can has to see there like here
8:17it it errors because it can't solve it
8:18because it doesn't have any modules
8:19whatsoever I just give it like a coroner
8:22module it can't solve it either because
8:23there's no way to build a Korean system
8:25with just a quarter module if I gave it
8:27corners I'm sky then yeah it kind of go
8:29the corner of course guy he's there but
8:31that's it so the way it sort of cuz I
8:37feel like it you can see the solving
8:39happening in a little tiny slider here
8:42and this is also available their own
8:44lines you can you can open this as I
8:50click a peak so the core idea is that
8:52yeah you have this huge possibility
8:54space where everything is possible
8:55everywhere and then as soon as you place
8:57something what you're doing is not
8:58really placing something of what you're
9:00doing is you're removing all other
9:02possibilities from that tile until
9:05there's only one remaining and as those
9:07are removing though they look like if
9:10you have one that has like sky downward
9:13pointing if then you remove that one and
9:16that was the last one that had sky
9:17downward pointing then in the slot below
9:21it it has to remove everything that
9:24relied on having sky downward so
9:26everything that has sky upward and then
9:28you know that might have further
9:29implications and has to remove some
9:31other things that rely on those things
9:32so since this is like I don't have any
9:35overhangs so as soon as I play something
9:38with dirt it knows that everything below
9:41has to be dirt right you see how they
9:42created this big dirt pillar here and I
9:44was gonna make a similar Twitter I'm in
9:47the tweet because they don't have
9:48anything but with in between there so
9:50that's a that's a recursive like I don't
9:54have to because obviously the first name
9:56approaching that's a completely fine way
9:58to start with it it's just like when you
9:59have made any change just check the
10:01consistency of every single tire like
10:03are there still adjacent possible
10:06modules that would satisfy like these
10:10kiss drinks that's perfectly fine way to
10:12start but then you can then you can
10:15build it in like a smarter way word like
10:16as soon as you remove something it
10:18checks
10:19was I had the last of my type that was
10:21removed and if I was then might let the
10:24slope next to me know that hey there's
10:26nothing matching this anymore
10:27and that goes okay that means I have to
10:29remove this one and they're like that
10:31propagates too so if we if we for
10:37example we go for
10:48so you see how I can paste this little
10:50so this all anchorperson so it's all
10:52implied from one of them and it just
10:54picks the first one so in this
11:01particular case the the because this is
11:04just a 2d picture so you can derive your
11:06adjacency constraints depending on you
11:10know what kind of tiles and modules
11:12you're working with in this case I'm
11:13just working with a picture picks a
11:15picture and it's not a very complex
11:17picture
11:18so I've derived that by just like you
11:20see if I click this is here some other
11:22things like up to so like if I click
11:25this one see this is like siding sky
11:27this is like top and bottom sky this is
11:32like sighing dirt top and bottom dirt
11:36and it derives those in this case just
11:40by sampling the texture on three points
11:42just getting what's the color here here
11:44here like along the top explicit color
11:46there there there getting the color just
11:49turn that into a hash and then you know
11:51everything that has the same color in
11:53those three spots will get the same hash
11:55and then you can index that too and then
11:57it's all very simple so so this the
12:00little numbers you see here it's
12:01actually better that color on that point
12:04in like hex values but that doesn't
12:06really matter
12:07I mean it's helpful if it's somewhat
12:08meetable to you is that you can compare
12:10oh I thought these pieces would fit
12:11together but then you can look all know
12:13the color is actually slightly different
12:15because they accidentally did something
12:16wrong of course you can round off those
12:18things before you put the diamond hashes
12:20to if you don't want to be too tedious
12:25yes yes so yeah
12:31the court and this is so this is an
12:33example of there's a bigger field of
12:35programming that I didn't know when I
12:36got started with this called
12:38constraints based programming and it's
12:42sort of the idea that instead of you're
12:45writing an algorithm by it you're
12:47starting with just declaring what are
12:49the constraints here like what things
12:51are allowed together with which things
12:53in this case of course is strictly
12:56adjacency constraints so it's just what
12:58things aren't allowed to what things and
13:00then like in all constraint based
13:02programming you just like asked how
13:04we're going to give you a solution that
13:05satisfies those constraints you have two
13:07constraints which like hard rules and
13:09then you can also have heuristics which
13:11I like ok but when there are multiple
13:13options right like here it doesn't know
13:15what to do because obviously like it can
13:16put anything anywhere but then you can
13:19have heuristics that help like oh you
13:21know high probabilities you can have
13:22this random picking with noise and
13:23things like that too but yeah so this is
13:26one example of constraint based
13:28programming which is a very useful way
13:32of writing code and migrating pools in
13:34in general so that's the simplified
13:37version I also made a little go there we
13:41go I made a little just schematic to
13:44show like an even simpler version this
13:46one doesn't move unfortunately it's just
13:47a picture but you can see how like here
13:50I just have two types of sites I have 0
13:53& 1 & 0 is grass and one is Road so I
13:58have two slots with a very very small
14:00system and each in each of these slots
14:02four modules are possible and here we
14:04can see how they rely on each other
14:06right so how like the this road piece
14:09this piece here and this piece here both
14:11rely on this little piece here right so
14:15if I would remove that piece then like
14:18this mine gets cut and there's nothing
14:21feeding into this anymore so then it
14:23tells the next one okay you now need to
14:25remove this one and this one and I'm
14:27like those two
14:29that it would collapse that so yeah big
14:32possibility space and then collapse down
14:34and resolving implications of your
14:37constraints and then I'll show you how
14:41this actually works in our game no
14:46that's something else there we go
14:50so this is five mark again we're making
14:52and I'll show you I'll go ahead and
14:55generate some new levels it's like a
14:59campaign map that now I have turned on
15:00to you so I can just like also generate
15:02whatever I want here so I'll pick a
15:04level and the generation is going to
15:08happen so here you can see he's just
15:10like the wireframes of all the modules
15:12that would be possible on all these in
15:16all these slots like it's much it's much
15:19slower now by the way because this
15:20visualizing it has its within so you see
15:26it's sort of building a possibility
15:27spaces first right because it needs to
15:28do that needs to propagate everything
15:30everywhere and then it needs to resolve
15:32initial initial constraints cuz the
15:36previous one we saw it didn't have any
15:37additional constraints anything can go
15:38anywhere but here like I need to have
15:40sky on the outside the level can all be
15:43inside dirt right and I also it's island
15:46so I need to have water at the bottom so
15:48that like initially implies the bunch of
15:51constraints you have a bunch of things
15:52being removed right here somewhere
15:55around there yeah now we've resolved the
15:57basics the basic initial constraint so
16:00you can already like start to see like
16:02the shadow of the shape of the island
16:06here you have the first thing actually
16:08being placed it's it's it's a house it's
16:11a hole for a house because the house is
16:13did I've added later because they're not
16:14part of the Train and you see how
16:17already like some options around the
16:19south side of the house started
16:20disappearing because obviously there's
16:22already
16:24well they did but a little bit further
16:26and he seems like having youth placing
16:29new things and resolving constraints so
16:32like here it started adding a niche and
16:34there we seem like the shadows like now
16:36there's a gap here because there's
16:37nothing like that would be possible to
16:38place today that would but wouldn't
16:42conflict with this edge here so you see
16:44these sort of this sort of ghost to
16:46represent yeah and they're now it's all
16:47generator so it started like acid gets
16:50going into its faster and faster and
16:52faster because most of the implications
16:53are already been resolved nobody never
16:59look oh I think that's a different yeah
17:01but this is like what the result yeah
17:12see seeds getting constructed so
17:14building spaces spending that's what in
17:18slows parts and I was getting really
17:21slow like placing all these cuz there's
17:23no need to be like them I think around
17:24like a hundred thousand possibilities
17:26just this light guy I don't there's a
17:28few hundred modules and then they place
17:31one on each tile on might rotate to the
17:34mirror to
17:53now it's starting to collect it it's a
17:55bigger level yeah and of course that's
17:57with a lot of procedural things the
17:59generation time increases exponentially
18:02with sizes are the bigger the bigger
18:04than that the slower it is though we
18:08always be solved that if I play on small
18:09islands so yeah it's starting to shrink
18:20it like you see some shadows disappear
18:22here that my Chronicle ologist by the
18:23initial initial implications
18:38holies yeah and now it starts like some
18:41of the initial ones and start placing
18:43placing tiles and this is also really
18:45like one thing that's interesting a big
18:48problem with procedural generation in
18:50general and with this technique in
18:52particular this like sense the technique
18:54the core of the tech doesn't know what
18:56we are generating but I need to generate
18:59playable levels right like where you can
19:01actually go all the places where you
19:02have to go and that actually took me a
19:04while to figure out how to do that in a
19:06good way obviously I can after i
19:07generated a level i can go through it
19:10and like validate mike has like can you
19:12go from the beach up to the house
19:14everywhere where you're supposed to that
19:16works but obviously you can end up
19:18generating a bunch of levels that you
19:20just sort of have to throw away oh look
19:23at that
19:26no no no it's a whole there's a cave
19:29inside but we won't be able to see it
19:31okay yes another level for us it hides
19:41them when they generate them because
19:42this is like I suppose happened sort of
19:44offline
19:48but yeah but so the way we did with
19:50navigability is you see these green
19:52lines here like the modules have a very
19:54simplified idea of how you can navigate
19:59from them so this is like a flat area so
20:01it knows that it can go to adjacent
20:03modules so when when I generate I this
20:08is like the heuristics I was talking
20:10about before so when I get to because
20:12sometimes it's just implied and you just
20:14need to place the module somewhere
20:15because nothing else is low but
20:16sometimes you you get to pick like
20:18especially when you start with an empty
20:20canvas you get to pick which modules
20:21you're supposed to place where and that
20:24then I drive this with like a
20:25navigability heuristic so I start with
20:27one place that I know it's navigable I
20:29start with a house and then it tries it
20:33tries to pick tiles that expand that
20:36navigable island so you see how it keeps
20:40like building with these sort of these
20:42green lines outwards
20:44there we go and here actually it starts
20:47here we can see it is complaining that
20:49these levels don't have enough beach so
20:52we've discarded them and tried again so
20:53we have to do that few times that's also
20:56an interesting question of like
20:56backtracking in generation in general
20:58like would you when you make an error
21:00when you run into a contradiction do you
21:02go back and then try something else or
21:04do you I tried that a bunch of times but
21:06the thing is that you don't know how far
21:07back you have to go you don't know if
21:09you like maybe the error was implied in
21:12the first pylons you placed so in my
21:15case I solved it by just I say the
21:17initial state when it's resolved all the
21:19initial contradictions and then we just
21:21if something fails if it runs into a
21:23contradiction it just reload reload
21:25stack and collapse again this way faster
21:27than symmetric
21:46they should built this visualization
21:48like yesterday so quite happy with how I
21:52can get sort of shadowy view of so you
21:55sort of see how the way they imply
22:00possibilities and you see you can tell
22:04also that of course like keep hearing
22:06the corners there are way fewer
22:08possibilities even in the beginning
22:10because there can like there can be a
22:11corner corner beach there can be a
22:13little rough in the water
22:14and it can be water but that's basically
22:16it and that's all like that gets implied
22:20by by just by the initial contradictions
22:25this is a taller one so I guess yeah
22:27here we can see the these sort of the
22:30navigable network being a little bit
22:32more interesting we should be able to
22:33once you start building that work yeah
22:36you see like here it's sort of a little
22:38stair going on here so there's not a
22:40bubble network goes up like this and
22:42then it goes down right so it keeps and
22:44this reading books like encapsulate the
22:48navigable network so the way the way I
22:51do this time - both heuristic is like if
22:53the new tile would increase the size of
22:56a navigable box then that gets high
22:58score and I think that one
23:05oh yeah and we can have a look at all
23:11the pieces I'm using for this so this is
23:13like all the tires I'm using unshaded
23:18some wireframe on them so it actually
23:21spent like now there's a ton of them I
23:22concur just about like 300 so that's a
23:24lot but it just started out with like
23:27maybe these ones something like that
23:29because this algorithm as I said says it
23:31solves the constraints it because with a
23:34lot of companies kind of combinatorial
23:36things you need to create a lot of
23:37different permutations which is the pain
23:40but since this one is driven from the
23:42competence instead of the content being
23:45placed to sort of satisfy the the sort
23:50of higher level abstraction so for
23:51example voxels or something like that
23:53you can get it going with very very few
23:55things it will just figure out how to
23:56build structures from that so I started
23:59with something like these ones just to
24:01get a sense of right if they're really
24:02important things like what's the like
24:04what do my hedges look like what do my
24:06slopes look like what's the corner like
24:08how much detail do I want right because
24:10it's a really important part but then
24:13you know it's really easy to just keep
24:15adding new content and you can choose
24:17yourself how many like how much
24:19permutations you want to add like these
24:21are all of my sort of little stair cases
24:23which like obviously you need a ton of
24:25those for all kinds of situations but
24:28there are some other things that became
24:29sort of occur on just sort of one-off
24:31locations like this little rock here it
24:33just goes in the water it just goes into
24:36work I also added a little tool so you
24:39can check adjacency so here is if I just
24:43highlight the sites of things other
24:45things get highlighted that fit with
24:47that thing so this will type it here of
24:50course fits with all the other little
24:52side bits that's really useful when you
24:55alter you Compton's and you need to
24:56check the hilts and you think this
24:57actually fit with the lesson thanks I be
25:00able do they actually function burkas in
25:02maya and not like it's also a little bit
25:04in the fours
25:06yeah yeah this is one maybe yeah that's
25:09uh that's cuz I really detest like
25:11keeping a good structure in my mind
25:14since things like that so what I do is
25:17like on when I import the FBX to unity
25:20I just merge it all into one and then I
25:22go through and cut it up into little
25:23pieces and analyze the edges of it and
25:26and it can also make bigger pieces this
25:29is quite an important aspect that you
25:33can make right so this piece here
25:35stretches across two tiles which is
25:37that's a super useful thing to use just
25:40to sort of break up the obvious tiling
25:42of course so I still like I don't hide
25:44the time and you can see in the game
25:45it's clearly a Titleist game but you can
25:48still like just like sweep around
25:49corners a little bit more like those
25:51things are really nice so that you keep
25:53them on their toes they don't know
25:54exactly what's going on it's really good
25:57for those things and it's also written
25:58of course crucial for making visible
26:01staircases because they have to make it
26:04have to go from one level to another one
26:06so almost beginning back here you can
26:09see also these little green lines here
26:11is the sort of pieces of navigable
26:14network that it's using for that with
26:18the navigability heuristic that I used
26:21to explain it and the weight the weight
26:23just two minutes when I import them to
26:26cut it up into pieces the weight figure
26:27self which one are supposed to go into
26:29small pieces and which ones are supposed
26:31to go into big pieces and actually check
26:33the plate does this have a triangle that
26:35spans over to another tile and then it
26:39just merges those into what so it's all
26:42in how I model them so I need to place
26:44all the mic it doesn't actually cut the
26:47mesh it just takes all the triangles and
26:49then puts them into different titles and
26:51then like figures out if some of them go
26:52together and then they get merged which
26:56is super I have some more examples so
26:57just like those sort of grounding right
27:00like this a little bit here just past
27:03like just rounds all finacial Oh
27:05which makes this a bit nicer and this
27:08little playing here actually move the
27:11together Middleburg picks it's just
27:12moved to the side a little bit good this
27:14is your trick the thinking - thinking
27:15that this is just one big plane instead
27:17of kiting up in the little place but
27:19yeah and so one thing artistically I'm
27:22quite happy with you if you look at a
27:24corner it's slightly asymmetrical it's
27:28like this side here is a little bit
27:30smaller than this side here that's a
27:33very easy way to get a tiny bit of
27:36variation so like on top of like a
27:38pillar that if you take four different
27:39corners of some of them are mirror then
27:41like there are a bunch of different ways
27:43that those could look like and also it
27:46means I can add these little nice pieces
27:49which are sort of adapters but where I
27:51can have like a bit of a crack which is
27:53like it is a detail but it's also sort
27:55of meaningful in the way that it it
27:57flips the big mirrors orientation right
28:00so it's thick here to thin there it's in
28:02there but thank you and that's I think
28:04that's a nice way of thinking about
28:06detail in general like making detail
28:08this that's meaningful and that DS
28:10changing thing have a look at a look at
28:16generated level here
28:19yeah I'm dimension also something about
28:21the history speed of oh I mean there's a
28:26lot of technical things going on here
28:28but there's also a bunch of the stage
28:29things going on here
28:30and you can tell that the like the
28:34interesting thing is not what happens
28:38with like on surfaces the interesting
28:41thing is what happens where surfaces
28:43meet each other right like if I would
28:45just kind of like a super obvious tiling
28:48grass texture here that wouldn't like
28:50that doesn't really provide information
28:51right we sort of yeah we get it's grass
28:53but like so the stories instead
28:56additional stories are being told where
28:58surfaces meet each other right so the
29:01first weeding obviously if it's like the
29:03grass patches like where are those
29:04levels and then we have the little edge
29:06you got here between the grass and the
29:09and the clips and then you have another
29:12little edge word about the
29:13convex where you put these little brass
29:16pins which also might break up the
29:18timing know that those are procedurally
29:19in place of course the grass patches and
29:24then you get these little sort of yeah
29:27here you can see one of these still we
29:28were adapters being used just like so
29:31you have a yellow pine like normal
29:32normally the corners just align but then
29:34sometimes it shifts and there's a little
29:36bit of a crack in it so that's a good
29:38gear to like soft line but a little bump
29:41in it listen I think that's a nice
29:43approach to detail in general you don't
29:45eat detail everywhere you need detail to
29:47smell to tell small stories and usually
29:50making transitions between things that's
29:52what details should do typically yeah in
30:00fact sort of most of the anti of course
30:04keep spying in too because if we've put
30:07some units on here it's also the idea of
30:13right where the environment is it is a
30:15cannabis for the gameplay so the
30:17environment like the emergency be
30:19beautiful but it shouldn't have that
30:21many details because the details should
30:24come from the game terrain and it's the
30:27same it's also but these units because
30:28in this particular game of course you
30:30control small armies instead of
30:32individual units so each of the little
30:35each of an individual units is quite
30:38simple in its aesthetics to sort of and
30:41the focus is being put on how they move
30:43as a unit in stacks
30:47and of course the the items are also
30:51quite literally a canvas for the game
30:53played in that you do if they do get
30:54painted might have a little blood yep
31:01that sort of it's about yeah it the game
31:08isn't completely frame rates independent
31:10unfortunately because I didn't know that
31:11much when I built it certainly so when
31:14it's a bit of a lower frame rate yeah so
31:20that was it for my presentations
31:26[Applause]
31:33[Applause]
31:41like once they showed through the
31:44beginning I've tried triangles on the
31:45sphere and that's what I've tried
31:51something that I've been very that I've
31:54really been wanting to try is doing sort
31:58of sort of cubish so like X cubes and
32:03the modules are cubes but you distort
32:05them to fit together so sometimes you
32:07have a 5 quart like 5 cubes meeting more
32:10corners sometimes you have three keeps
32:11meeting in one corner I think you could
32:12and then you like just distort the tiles
32:14to fit them in there I think you could
32:16make really beautiful sort of medieval
32:18seats and things like that with my
32:19curving roads like that with the system
32:22like that just I I find when you build
32:26different tile structures sort of work
32:29for different things right the triangles
32:31so that's hexa bombs right because
32:33triangles and hexagons are related are
32:34really good when you make like for like
32:38the planet I showed first for example
32:40when you have bigger terrain shapes and
32:42then like face they're decent or two but
32:45four houses in smaller shape so it's
32:47quite hard to do with triangles it's
32:48much easier to do with with cuse because
32:51that's sort of good structure to shape
32:52yeah but yeah you can do it with
32:54whatever rigid one you don't even have
32:56to do with a grave you can do it with my
32:58completely arbitrary graph and you can
33:02mix it right you can use both triangular
33:03tiles on just like tiles some cute eyes
33:06you've two dimensional of course
33:13the example you showed that you choose
33:21yes how did the game choose which
33:24direction to go yep so that's that's
33:28approaching general is called the mixed
33:29initiative approach we're like you asked
33:32user of some initiative and then the
33:33algorithm has some initiative
33:34the game isint makes finished it at all
33:36like that's all just being picked
33:39randomly both because of course as you
33:41saw in the example there were like
33:43sixteen options in the game there's like
33:45300 options so like how would you
33:47present that I'm sure there is a good
33:49way to make the proper intro having I'm
33:51sure there's a really good proper
33:53interactive tool in this but I haven't
33:54built it so no the game the game just
33:59like picks them according to different
34:01rules
34:02it starts with houses for example
34:04because the game the game relies on
34:06there being houses and also I have a
34:10bunch of things to drive it so I have
34:11like fiends and some boxes around these
34:15which like have names and they have
34:17different rules so some things for
34:19example have rules that they're not
34:21allowed to be next to themselves like
34:23copies of themselves quite an important
34:25rule is for modules I don't want to be
34:28next to a mirror of themselves because
34:30that usually looks quite bad when you
34:31have generic pieces next to each other
34:33and how comes quite often because of
34:34course a piece fits perfectly next to
34:36the mirror of itself but I also use
34:39these four themes right so I have some
34:41like have some cave parts here this is
34:44like the case which I only use for some
34:47levels and then I have some like so my
34:51ruining structures kind of
34:59there we go yeah it's this is not my
35:05normal work station but yeah I have some
35:09like Ruben type shapes too that I use
35:11for different than I then I have some
35:12with like some big boulders and rocks
35:14and things like that this is actually
35:15this level here's is a rock theme right
35:18so you see I have this little nice
35:20little overhang here it's just like a
35:24bridge so I had some themes like that
35:29just to keep you I mean that's also
35:30always a challenge with procedural
35:31generation that think things can look to
35:33see me and this is one of the ways I
35:36approach that I mean obviously also
35:37switching colors is searching foliage
35:39things like that also yeah
35:54no I'm sure that's good
35:58I'm sure there's a lot to be gained
35:59there it's everything so little pull
36:01here though so modulated myself this I
36:04mean like it's just a few triangles yeah
36:14actually you said like for one of the
36:16island the algorithm has to restart
36:18twice because the you find enough yeah
36:21but when do generic can can you like do
36:24you do you shift the weights of the
36:26different times you're gonna put or are
36:28you are doing this the same just they're
36:30just you're straying from the other I
36:33mean there's both forming the observers
36:35the adjacency constraints that have hard
36:38constraint right always but then there's
36:40also probabilities and I can put boxes
36:42around things and see like hey the tiles
36:44within this area nation and I was like
36:46the higher or lower probability so yeah
36:50but the beach thing that happens
36:53afterwards when it's like it
36:55this algorithm is done and then it's
36:57constructing the navmesh and then it's
36:59checking like is there enough beaches
37:01where ships good land if there's not
37:03enough than described gave in this time
37:05but it kind of fail very earlier in the
37:06back to and it can fail like there are a
37:09few validation sets after generation and
37:11there are a few during so if it closes
37:14the navigable island so the the network
37:17its building if that's if that gets
37:20closed before it reaches the bottom for
37:22example then you know there's no way
37:24this
37:25and then I can discard it before I do
37:32you

About this transcript

This page contains the full transcript of EPC2018 | Wave Function Collapse in Bad North | Oskar Stalberg by BUas Games, generated from the public captions YouTube serves with the video. The transcript has 6,335 words across 880 segments, with the original timestamps preserved so you can click any line to jump to that moment in the embedded player.

What you can do with it

Use the transcript to take notes, quote the speaker, build a study guide, generate a summary with ChatGPT or Claude via the YouTube Summary tool, or export it as a timed subtitle file with YouTube to SRT. You can also re-open it in the transcriber to translate the transcript into 100+ languages.

Free YouTube transcript tool

YouTube2Text is a free YouTube transcript generator — no signup, no daily limit. Paste any YouTube link and get the full transcript instantly, with timestamps, click-to-jump, translation to 100+ languages, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.