Procedural Level Design in Eldritch

[Music] all right I think I'm going to get started now um before we begin the usual reminders please silence your cell phones and they're doing the the P scanning thing oh wow they're doing the P scanning thing so between each session make sure that you get your get your pass scanned so that when you do your evaluations you get one for each of the sessions in the talk so all right uh hi thank you for being here I'm David Pitman I'm going to be talking about procedural level generation today and I'm just very briefly going to introduce myself it's my first time speaking here at level design in a day and my background is a little bit different I was not a level designer before I I used to work in big budget games especially at 2K Marin where I was an AI programmer on BioShock 2 and for a time I was the lead AI programmer on the what eventually became the bureau XCOM to classified I kind of got burnt out on that game and about two years ago I had the opportunity to do something new

I co-founded minor key games with my twin brother and now I do a little bit of everything I do programming design bad art audio business and marketing pretty much anything that has to happen to to ship a game I do it now so about two years ago I made this game called eldrich and when I was working on it I actually had a kind of difficult time explain to friends and family what I was making it wasn't like a particularly Innovative game it was just a hybrid of genres that people didn't really visualize the way that I was thinking of it it was kind of hard to describe I tended to talk about it in terms of my influences in immersive simulations and Rogue likes but what I was really trying to say is if you take first person combat and magic and stealth and you mix it with procedural levels and permanent player death and dress It Up In Love crafty and fiction and make it on a Sho string budget you get something that looks like eldrich so minor key is a two person

company it's myself and my twin brother Kyle we actually work on individual projects though I made Aldrich completely by myself in a span of slightly less than eight months which I don't track hours I think that's silly especially for independent developers but back of the envelope calculation puts that around 1,500 person hours of work to make this game that number didn't mean a lot to me I so I thought about like what would it be for like a conservative big budget game that would be maybe 100 people working for 2 years and not crunching that would be like 400,000 person hours so this was like a fraction of a percent of the work of a large game and I'm not saying that like to brag like the game has a ton of rough edges but it's just I had a very fast or a very short time span to make this game and I had to work very fast and so that's going to be something that comes up throughout this talk is trying to find the simplest thing that could possibly work and that was actually why I chose to do procedural level generation for this game um despite the the kind of rushed development of eldrich it was fairly well received I've cherry-picked the ratings I show here

because it's my talk and I can do that U but you know nine out of 10 people who wrote A a review on steam liked the game so that was pretty cool and it made a tenfold return on investment which was also really nice and it didn't do that by selling like millions of units or millions of dollars like it was not that successful but the development was so lean that it was fairly easy to recoup the cost so I want to talk about specifically what I was trying to accomplish with the levels in eldrich but first I want to talk about why why why I used procedural level design for this game and why anyone might use it so the first reason is probably the most familiar is if if the game calls for like an element of perennial Discovery the way that Rogue likee games do every time that you play a Rog likee or some strategy games like Civilization uses randomized maps to great effect the player can't go into the game knowing what that level is going to look like they can learn the systems but they can't learn the space and so the the game design in that case actually dictates that you need a generator so that the levels are different every

time project constraints are another big one and I think this is the reason that there's you know kind of the current invogue Rog likee scene amongst independent developers myself included when so as an individual developer I didn't have time in eight months to make even like five to six hours of gameplay if I were building the maps by hand I was trying to do everything else too make enemy AIS build the physics for the game like I I wrote my own engine for this game it was I I I just wouldn't have had time to do love if you're familiar with the uh like the project management triangle of cost time and scope where if as one of these things changes the the other two have to push a pull to accommodate it like as an individual developer cost and time are the same thing like this triangle basically flattens into a line between how much money do I have and how big is the scope of the game going to be I had to keep the scope of the game small because I was going to run out money I couldn't have continued to do that as a full-time job

sorry another reason to consider procedural level design is if it's you know if it plays to your team strengths my background was as a programmer I was able to get very fast at developing uh procedural levels I had done this sort of thing just as a hobby before and so when I went to make this game it was natural for me to want to do this whereas if I were trying to make Levels by hand I'm not an experienced level designer I would have been kind of fumbling at it I would have been much slower about it uh procedural levels also kind of enable content reuse in a really good way everything that I made for this game kind of just ended up in the mix to where if I if I built a module I I made some new art like it just it ended up showing up everywhere in the game you would see it regularly you would play you know the the structure of a rogue like is such that you die you play a level over and over you know A variation on the same level and so you saw all the content in eldrich a lot there wasn't the kind of in in uh big budget games especially there there's a lot of money spent on set these things that you might

only see one time in 10 or 12 hours of play there's multiple ways to use procedural level generation the the one that we're familiar with the one that I use for eldrich is generating levels at runtime but you could also generate levels offline and then work on them by you know like I I can imagine this is a good way to as as an indie developer create a game that's a bigger scope than what it looks like your team could support like generate a large terrain randomly and then fill in like key points of by hand you could do something like that you and you don't also have to make a whole game that's based on random levels you could supplement like a 10hour campaign that's built by hand and then if the player wants to do more after that like there's an endless amount of generated levels so they could play after that I actually don't know of any games have done that but it seems like it would be a really cool thing the last point I want to make is that procedural level generation is design I think there's a tendency to regard it as the domain

engineering but I can imagine kind of like a hypothetical programmer who doesn't have any design sensibility that person Couldn't Write a good algorithm to generate levels because they don't know how to evaluate the output of the system like you have to know what's a good level in a way it's almost like you know if you're a lead level designer you may not be the one building levels but you have to be able to tell the rest of your team you know here here's what the levels for this game are going to be like and be able to recognize if something isn't fitting what the game needs that's kind of the same thing that working on procedural levels is is that the uh excuse me fight in the cold you need to be able to develop a a set of roles that will produce good Levels by knowing how to by knowing what a good level is and that that's why I think this is important to to talk about at level design in a day is because you know this is a room full of people who have internalized what is a good level you have expert knowledge you are used to building levels you know I'm I'm kind

of fumbling around in the dark like I'm not an experienced level designer but I hope to share you know what I learned along the way and show how I did you know take some design sensibility and shape what what I made with eldrich so the specific goals I had for levels in ELD I I was not trying to create like the sprawling random terrain of something like Minecraft I wanted to do compact levels that have had a clear start a clear end they took 2 to 5 minutes to go through I wanted the spaces to look I put handcrafted on the slide that's actually not a good word I wanted them to be synthetic worlds look like they were constructed I wanted to have ancient sunken cities and not just like terrain and cave systems I couldn't use noise I had to do things that were uh hand handbuilt you know I want the the effect to be of continuous space instead of room to room there's a few contemporary firstperson Rogue likes on

the market they tend to be more like Arena Shooters you get locked into a room you fight Monsters the door opens you move to next room and that's great like I really enjoy a lot those games but that wasn't what I was trying to make with Eldridge and the last goal that I had here was I wanted the worlds to be semicoherent which is a kind of strange goal normally you would want worlds to be coherent you want like a kind of internal logic and consistency but I was I was looking at Lovecraft reference I was reading Lovecraft story and that was sort of the inspiration for the world and one of these things that comes up in a lot of his writing he he describes environments in detail but there's a kind of sense of alien wrongness to the worlds that he describes this is from K cthulu he talked of his dreams in a strangely poetic fashion making me see with terrible vividness The Damp cyclopian city of slimy greenstone whose geometry he oddly said was all wrong it's a recurring element in Lovecraft and one of the thing I mean sometimes it takes the form like nonukan geometry and actually like three-dimensional space kind of being in Warped in a way that

our minds can't comprehend in eldrich I wasn't interested in doing that sort of thing but I want the idea that the worlds are not logical not consistent something's wrong about them and that's something that I don't actually think I was very successful at but I wanted to put it out there I mean that's kind of what GDC is about talk about your mistakes as well I was also looked at Art reference things that were specifically based on lovecrafts writing as well as you know just pieces of art I I've seen and think you know what this is giovan Batista paranis ceri wood it looks like a video game environment right and the stuff like I was getting from the the art references like I wanted to have right angles you know big cyclopian masonry really complex layouts like if you look at ceri like you see these staircases you've got catwalks you've got people all over the place but the connectivity of the space isn't really clear at first glance that was kind of what I wanted I want the sense of verticality that's not common in first person shooters that ended up kind

of coming back to bite me later I'll talk about that and I wanted I mean it's not about the structure and the layout specifically but I wanted to use light and fog to create a sense of atmosphere that makes these relatively small levels feel bigger than they are and the results in the best case are something like this we've got like if you know what you're looking for here you can kind of see the edges we've got on the on the right side of the screen we've got like green grass and then there's a kind of clear divide into a module that's made out of stone but it's continuous space it's not room to room we've got things that look familiar we've got ladders we've got scones with lights but it's only semicoherent like we've got a ladder that goes up to a stone catwalk it's not really clear what that's for that was kind of what the best I could do in the time I had for that sense of alien wrongness I don't think it was successful but these are the kind of levels that we get from eldrid so to give you an overview of of how the levels were generated there's kind of three three steps we start by building threedimensional maze just kind

of in the abstract and I'm visualizing here in two Dimensions because that's way easier to put on a slide but if you imagine you you've got a box it's sliced up along every axis evenly or like if you ever had like the clear plastic maze toys that you have to get a ball bearing through you can visualize it like that so we we build the Maze and it defines the high level structure of a level but it doesn't have any geometry it doesn't have any environment art then we populate it with a modular kit of rooms and this is the sort of thing that you might even be doing as a level designer now if you have like you know modular environment art and you can snap pieces together it's basically exactly the same concept and then we generate entities enemies and items I just want to give credit where it's due also this is not like a particularly revolutionary idea I was inspired in large part by the Spelunky generator uh the so Derek U who made Spelunky has written a bit about it but Darius Kaz Amy kind of separately re implemented the whole game in Javas script I believe and has made like an

interactive version of Spelunky generator that you can play with on the website it's really cool like check that out if you get a chance so the first step we we have to generate this Maze and you might wonder why we start with a maze because like if I'm playing a game and I say yeah it was really cool it had some good things some bad things I played this one level and it was kind of maze like do you think I meant that that was a good love or not like maze has connotations about I I probably meant that that was you know difficult to navigate I wasn't sure where I was going maybe I got lost maybe it all looked the same like Maes to me say I don't know where I was going and the the environment was all kind of homogeneous so why would I start with a maze then if if there are those negative connotations well for a Rog likee game not knowing where you're going that's actually a good thing that's part of the central mechanic of a Rog like is there there isn't like explicit way finding like there's not a designer who's who's trying to guide the player to the exit part of a rogue like is you have to find

the exit so that part I was just okay with and I actually wanted to have really twisty confusing levels because it did come back to that kind of love crafty and alien wrongness if you have like a confusing level and you get lost in it that's actually kind of cool that's part of the the Lovecraft experience but what I didn't want was the the idea that the environments are all very homogeneous like every part of a level looks like every other part that is not a good experience that that helps you get lost if you don't have visual landmarks and things that make each moment of a level memorable so I want the structure of a maze but not the visual sameness of a maze so there's a lot of ways that you can generate mazes I think the extent of research that I did was to go to the Wikipedia page for Maze generation and pick the first thing that looked like I could use it I ended up with something called prims algorithm Ms liked it uh I ended up with something called Prim algorithm which the steps are described there but we're just going to go through it visually but basically you you start with the Maze and it's all closed off to

begin with uh basically it's a bunch of rooms they have walls between them and they're all we call them closed and then we're going to pick one at random we're going to start in the lower left we're going to say all right that's the first room in our open set of the Maze and then we just randomly expand to any neighbor of that cell so maybe we we go up from there then on the next step we're going to pick randomly from the open rooms and expand to any closed neighbor of that so we go that way and we keep doing this and we keep growing until either we've we've done enough passes that we're happy with the maze or eventually we fill all the space so if we if we consider like the structure of this maze like it's it's a decent maze it it looks balanced like a maze should be and as much as a maze can have balance but there's no Loops in it that's kind of a property of the algorithm like because we open a path from a a open node to a closed node we can never join two open nodes together and create Loops there's also no open space in it and those are things that I

I ended up trying to find ways to fix so I made a few modifications to Prim algorithm there's two random factors in in prim's algorithm there's randomly pick a node to expand from and randomly pick the node that you're going to expand to so I biased which node that I'm picking to expand from based on trying to expand from the most recent visited room if I crank that knob up this was a this wasn't like a constant I do it all the time this was a a control that I had to try to differentiate iate worlds if I turn that knob up basically I get like the the generator kind of just expands through the space until it terminates in a dead end or it has nowhere left to go and so you get long twisty mazes that end up being kind of claustrophobic to explore they don't have a lot of branches in them if I turn that knob back down I get branches from anywhere in the Maze and it's more open space lots of Junctions less twisty claustrophobic uh spaces so the other random factor is which room we're expanding to and I biased that toward expanding on the horizontal plane because remember we we're doing this maze in

threedimensional space it it doesn't look like this it looks like kind of you've got three levels of this stacked up in addition to opening paths north south east or west we can open paths up and down but people are used to walking in the ground plane and we're you know even as firstperson game players we're we're trained to navigate in that plane and vertical traversal ends up being really difficult even in a game like elrich where there were a lot of powers and abilities to try to give you vertical traversal it just wasn't a lot of fun to do so I I biased the the maze expansion to always explore the ground plane and only when it terminated and it couldn't go anywhere else then I would make a path up or down and continue to explore kind of the next story of the Maze and the last modification I made to prim's algorithm it's not actually even part of the algorithm per se but after we've run this all these steps and we we've generated the maze I just randomly knocked down a bunch more walls and what that gets us is you know we look at this and we wanted to have Loops in here I wanted to have some long lines of sight

some open courtyards and and the Maze Generation by itself wasn't producing any of that but if I just start knocking out walls I get this sort of thing and I could have done a really intelligent system here of like you know try to identify a long Loop and find like the wall I can knock down to make the most interesting kind of level it turned out in a maze this small knocking down any random wall gets me exactly what I want and never has any bad side effects so it was like the again the simplest thing that could possibly work and that yeah it took me like 5 minutes to write and it makes a huge difference this is actually the modification to prims that that makes the most difference in a level because if I turn this all the way up like I get wide wide open spaces where if I turn it down I get like claustrophobic levels and it's the the the difference in like it starts from the same maze algorithm but you end up you know in a a picture like uh sorry I got to go way back so in a

picture like this it's a very open space we've got like connectivity all between here a maze by itself could never produce this right because there would be walls including all this so yeah this ends up being the most useful factor of the three modifications I made sorry so that everything I was talking about with the maze generation so far I'm kind of talking about regarding all space kind of as equivalent but in a level in alrich there's there's other things in the the level besides just corridors and and Corners there's the entrance and exit to the level are kind of unique features there's shops there's Bank vaults to Rob shrines to acquire Powers at and each of those is going to have to have environment art and a module made for it and that has to fit somewhere into the maze shops you always from one

side uh shrines they I I would sometimes tend to put them in in Corners the exit of the maze it always had to go somewhere specific like the the the entrance had to go on the top of the maze the exit had to go on the bottom so I had these special rooms and they had to go certain places in the Maze and the the first way that I wanted to do this was to generate the Maze and then search it for where I could put these features the problem is because like the geometry of of the exit the geometry of a shop any of those kind of things it had fit somewhere specific in the Maze and there wasn't always a guarantee from generating a random maze that I would have a place that could go for example I want to put a shop into like a dead end because it has one entrance in this example there's only one place that could go in in the top right corner that's the only dead end in the Maze like if I kept cranking that that open tuning knob way up there wouldn't be anywhere to put a shop and then even though I need that like for balance and for tuning purposes it just couldn't fit anywhere so I found I could do I don't search the maze for a fit for these

rooms I actually seed the maze with the rooms that I need and I prescribe like this room with this structure has to go here in the Maze so it looks like this it starts even before we run prim's algorithm we've got the Maze and it's all closed off and we say all right here's going to be a shop here's going to be an exit and we open up not just those rooms but also their neighbors and then I lock the rooms that they actually have those features in them the shop and the exit and I say when I'm doing the maze expansion nothing else can ever touch those walls so I run the maze expansion and it goes from from the neighboring rooms I opened and we get the whole maze built out here and that I've got that guarantee that like nothing is going to knock down the walls of the shop even if I crank up the open up the world tuning knob all the way and knock down all the rest of the walls the generator is going to guarantee this shop is always going to have this structure so now we've got this kind of exract maze fill in threedimensional

space but it doesn't have any environment art it doesn't have any geometry yet I use a modular approach to this um like you may have heard talked about last year for Skyrim I build small chunks of the world and each cell in the Maze is going to get filled in with one of those chunks ELD uses a voxal engine the each module is 12 M by 12 M by 8 m high and I just build them kind of like the way you would do if you're in like creative mode in Minecraft I have an Game editor I put voils down the challenge is that every module has to connect with any other module I don't have any guarantees about like which ones are going to be picked adjacent to which others and with you know 12 meters on a face there's kind of a lot of ways there's a lot of space where they could join this was another case where I just did the simplest thing that could possibly work I made a safe region on the exterior face of each module and that's what's highlighted in yellow in these images so any module that has that intends to connect on that

face it has to have that space be open and so like in the image on the left this this module connects in both directions one face is completely open the other is Walled off except for that safe region that ensures that those two faces could join together it will always be traversible by the player or by enemy AIS I could get some kind of fun effects by playing around with opening up geometry in on the exterior face outside of the safe region in this example like this sorry this picture is kind of hard to see but we've got a corridor so we're looking at the module from the outside and we've got a corridor that goes inside in the safe region but we've also got kind of an opening up here in the top outside the safe region and what happens is that we get kind of just serendipitous Effects by when when this module meets up with other modules it just kind of reinterprets what that opening looks like so the typical like the most common case is that it just looks like a balcony open on opening up onto the adjacent module but sometimes there's some geometry in

the adjacent module it looks like this is part of a a feature you know that the the two rooms shared that's kind of crumbling away sometimes we get because the the voxal grid is so low resolution they're 1 meter Cub voxal we we just get a lot of features that sendly join here so in this the adjacent module had like an upstairs level and it just looks like it connects to that opening and actually this really hides like the boundary between the two modules really well like this just looks like one space you wouldn't look at this and assume those two separate modules and then sometimes because this opening is outside the safe space it just ends up running into a wall and the player you know they approach us from the other side they open the door and they're like looking at a dead end it's just a wall there and I was like well that's that's not good you know I I wanted the levels to be semicoherent but that just seems like a bad use of space and then I I realized I can just put some Loot on the ground there and then it looks like a loot closet and it looks completely intentional so yeah I was just this was the sort of thing and I didn't do this too much in the game because it did tend to feel uh like like

players started to perceive this pattern and it started to break down but but once in a while it was kind of nice to get these kind of serendipitous results so in the Maze there's there's six directions that that any room can exit north south east west up and down that means that there's 64 different combinations of room of of ways that rooms can be configured uh two to the power 6 64 that means that ostensibly we would have to build a module to fit in each of the 64 configurations but of course we think that wouldn't make sense like a room with an exit to the north it's the same as a room with an exit to the South just rotated so the generator will do the rotation for me so if we have a tile it's got exits to north south and east the generator can rotate that about the vertical axis like we would expect and and give us a module that will fit into the other rooms this is just what you would expect if you've worked with modules before it's just like laying down the tiles in the city building game but the other thing that we can do that you couldn't do with physical tiles is

we can also mirror this and so when I do this I get a North South and West exit which we already had in the first row from rotation so mirroring isn't necessary like it doesn't it doesn't fill any void in level generator but it does kind of reuse the modules in an interesting way because if you rotate a room like the player doesn't perceive that you know if we took this room and rotate it at 90° it just feels like the same room we're just facing a different direction and it when you mirror it you do get you know the flipped image and the player actually perceives that as a different kind of room it did add the constraints that I couldn't ever like I couldn't try to spell something out on the voxal because then I would get the mirror image and the word would get reversed so by using those those transforms we get 24 configurations in the Maze that we have to build modules for instead of 64 and if you know I can imagine if I were building like a six degrees of freedom descent style shootter I could even reduce that set more because you could take like a room that has an exit to the north and rotate it up but since eldrich has an up Vector

and gravity I couldn't do that so 24 configurations to build modules for that doesn't sound too bad but we don't want the maze to be homogeneous we want we don't want for example every T Junction to look like every other T Junction so I need to build multiple modules for each of these 24 configurations and that starts to sound like a bit more work I also don't know immediately how like how many modules do I need I want to avoid rep ition in the level I want to have a number of modules I can plug in to any similar configuration in the Maze but how many so I wrote a little tool that generates a maze 10,000 times and just dumps a spreadsheet of like here's how many times I had a dead end room here's how many times I had an L Junction or a t Junction and then it sorts those by by the frequency and so that doesn't give me like an absolute metric of how many modules I need to build but it gives me a comparative metric of like appear three times as often as T Junctions I

should build three times as many l Junction modules so they'll you know there's a bigger set to pull from because you'll see those more often in the Maze so I ended up just I mean this scales really well I could build like even today I could go back into eldrich and build more modules and it just gets like thrown into the mix and when I generate a level sometimes you'll see those sometimes you won't I just had to you know kind of decide arbitrarily where I was going to cut that off and how much work I was going to do what I did was like in the statistical report if an L Junction shows up eight times in a level I built eight variations of that module it just seemed like a good number to do and so between all the the modules that will go into the ma and modules that had specific features like shops and entrances and exits and shrines I ended up with about 60 modules I had to build for each world of the game and in in the first version of eldrich they shipped I think there were about 300 modules across the course of the

game so one of the things I wanted to do was have each world in the game represent kind of a one story from from lovecrafts U this the the first world of the game was Loosely based on the shadow of insouth and it's supposed to represent kind of a cave system and sunk in Ruins beneath a New England town and the way that I wanted to do that was to represent it like the the town with red brick and steel beams and then have a kind of cave system beneath that and then sunken ruins at the bottom and so I just started building modules that had each of those themes in them and it ended up being very visually noisy it it felt chaotic I wanted levels that were semicoherent but this just felt noisy it didn't it was beyond not making sense to just not looking good so the first thing that I thought I could do was well I'll just split each of those themes out into its own modular set I'll have first you'll go through the warehouse level then you'll go through the caves level and then you'll go through the through

the ruins level the problem is that would triple the amount of modules I needed to build each of those would have to be a complete set my solution here was I I just explicitly declared each module that was building as part of one of those themes and then when I generate a level the generator elects one of the themes from the set to be the dominant theme so it says like maybe this is going to be a warehouse level and it prefers to use Warehouse modules when they're available but if for example I'm building you the maze has some some room in it and I don't have a warehouse module that fits that room it just falls back to using whatever else that can fit in there and so then I go from like the scene on the left if I I do like the colar COA and pass over it like the bright green Parts here are the ruins level and then the dark blue is caves and the red part is Warehouses and the image on the left is what it looked like before where it's just you've got all these themes all in the player's view right here and it's just kind of visually noisy after doing this themes pass I ended up with levels like what you see on the right where everything that you can see inher is

from the ruins theme and then you know maybe at some part in the level I don't have a a ruins module that fits in there and so it falls back to using caves or warehouses and at that point it looks okay it's like sure I buy that maybe you know this this sunken ruin touches into a Cave System somewhere that that looks fine but it's not so visually noisy so most of the modules I've been talking about are just things that that populate anywhere in the Maze but I also need to have the module mod for feature rooms uh entrances and exits shops shrines all these things this is basically the compl of the prescribed rooms in the Maze the things where the geometry had to be locked down in fact in the data it's all the same thing like this is the definition for a bank vault in the first world the first like six lines describe where in the Maze it's going to go and then at the bottom it says these are the modules that it can use there and I use this for a lot things not just like the the shops and things but also to add points of kind of really rare visual interest things that

would be uh key landmarks kind of both visually and experientially uh the image on the left here is from the second world of the game and it's it's kind of hard to see probably but it's a throne room it's being guarded by these two statues they're like weeping angels from Doctor Who if you're familiar with that they come to life when you look away from them there's no other way to get to the throne room except to go past these guys so it becomes a very dangerous challenge this is the sort of thing it's memorable when you see this it stands out it's you know a really cool encounter that's a little different than just fighting enemies in the rest of the game but I didn't want to use these too much I didn't want to have them show up all the time so at most you get one of these per level there's not even you know guaranteed to be one but just let me do something you know once in a while just mix something fun and just a little you know kind of unique bespoke encounter that gets mixed into the level somewhere I also used them uh feature rooms to do uh things like the image on the right it's a zigurat at the end of the first world and it's bigger than any feature that you could that the maze could generate on it it own because it's actually built out of eight modules kind

of arranged into a giant Cube so you would never see a feature this large just in a typically generated level in the game it can only happen kind of when I build it as a bespoke thing so now we have we we've built the maze we've populated it with the modules and we need to spawn some some objects into it some enemies to fight some loot you know the goals I had for this first are kind of the things that you would expect building you know Levels by hand you have to challenge the player with enemies and traps you have to give them incentive to explore by having loot to pick up for a procedural game there were additional goals of like I have to actually place these in sensible locations or the the the generator has to place them in sensible locations and I also had to control the group population like I didn't want to leave it up to chance how many enemies are in the level I want to be able to ramp that up over the course of the game so like the first world there's a dozen enemies and at the end in within the same space there might be two dozen but I needed to be able to control that I never even considered trying to write

an algorithm to place entities like there's just so many roles like if if I'm trying to place a light in the world we all kind of have assumptions from being people about where a light fixture might go but trying to codify that into a rule set would be really really difficult so I knew that I wanted to place entities by hand but I didn't want to say like everywhere that this one module shows up there's always going to be this kind of trap or this kind of enemy um and I you know I think the sort of thing is used in uh games with handbuilt levels as well as procedural games but what I did was to just kind of abstract the concept of putting an entity into the world one step away so that when the when I generate the level there's a rule set that runs that says like all right here's a thing you know as a designer I I put a a spawner into the world and there's an opportunity for something to spawn there but then there's a r set that's going to run and say do I actually want to spawn something right now and if so what's it going to be so the most basic version of this is like a spawner that says it always spawns and it's always this one thing and we use that like for the player spawner because you always need a

player I could do spawners that had a random chance I did this a lot for traps I would you know just cover a module with like spikes on the ground or trip wires crisscrossing it but then give them like a one and two or one and three chance to spawn so like the the specific arrangement of of the traps in that room will be a little bit different every time that runs what's more interesting that I could do is to actually randomize What entity was spawned there this is an example of when you've broken into a bank vault you know there's there's a bunch of coins to gather up from behind behind the Gates and it has a 75% chance of spawning at all and then if it does spawn it's got a three to two chance to spawn a big pile of coins which is actually a separate entity than just a single coin I used this also for for enemies I would just place a generic enemy AI spawner and then it had a table of like here's all the enemies that can actually be for this level and they're weighted by like we want a whole bunch of kind of grunt enemies but we only want one really tough guy so I could do that kind of thing then I needed to control the

population I I called this resolve groups which is a kind of bad name but basically the idea was I gather up all like the spawners that are tagged a certain thing like all the enemy AI spawners and then I run some rules I call the set of those and I'm able to resolve that into a subset of things I actually want to spawn I use that for enemy AIS for a few really rare items like healing fountains that they they not only heal you but increase your max health things I really needed to control the number of those that appear throughout over the course of the entire game and this also let me run some additional rules while I was doing it like for enemies I want to enforce a minimum distance from the player spawn it was a really terrible experience early on in the game when like you would you know I never put an enemy AI Spawner in the same module as the player spawn it could end up like adjacent to something that had three enemies in it and suddenly you know you've got monsters in your face and you've just entered the level so I just put a minimum spawn on there so when it when it's resolving that big set of spawners into the small subset it's actually going to use it just gets rid of anything that's near the player

anyway spawner placement was something I didn't think a lot about until very late in development when I started I was just I was building modules and I put enemies in in them and I put some traps down I put some loot I was just kind of putting a little bit of everything everywhere and as as the game progressed and as I I you know play tested it and and felt more comfortable understanding what the game was I realized that the placement items was actually really key to guiding players through the level even in a randomly generated level players are really good at perceiving repeated patterns in games like it's really easy when you've played the game a few times to say like oh I recognize that that room is a dead end it's not going to get me to the exit I should just not go that way because I'm you know I'm doing fine I just want to get to the end of the game right now instead what I started doing was I I didn't put loot like along in any room that would have Junctions I started not putting loot in there and just putting it in dead ends so that when you're going through the game you do some

combat you're running low on items you see a dead end and you know that it's not the way to the exit but you also know there's going to be loot there you know that's where you can go and replenish your bullets replenish your money uh traps I ended up moving into hallways and I I use hallways kind of liberally here to mean any module that has two exits whether it's literally a hallway or an L-shaped corner or a vertical Min shift anything where the player's experience is really just to Traverse the space there's not another choice to be made there that was a good place to put trups just make that traversal a little bit more interesting I also put trups in dead ends kind of to guard the loot that was a really good place for them and then enemies I I wanted enemies to kind of congregate around Junctions there was no really good way to to ensure that because enemies and eldrich they wander randomly it kind of didn't matter where I spawned them they were going to wander all over the level wander into the traps get themselves killed but the the the target experience I had and it didn't quite work out was to have enemies at Junctions where the player could choose to you know circumvent an encounter or

sneak past it and have some options and use this you know the fact that there's multiple ways into the space as part of that encounter like I said this is stuff I didn't start thinking about until really late in eldri's development and it's kind of this is what I want to leave you with and this is the key takeaway is this stuff made the game way way better and this stuff is not like mutually exclusive from random level generation the the kind of rules that as level designers you've internalized you have your expert knowledge of what makes a good level and how to build that if I had had that I would have known to do this sort of stuff much earlier in the game I would have been thinking about these kind of things earlier instead of having to kind of stumble through it so that that's what I want to leave you with is I hope that as level designers you can be like super excited the way that I am about procedural level design but bring the expert knowledge that you have that people like I don't have into it and find you know ways to do generated levels better than what we've seen before you know make make generated

levels that are better than eldrich uh because you know what makes a good level and because you've got that experience and that practice so I think I actually blazed through this a lot faster than I intended so that's my talk uh thank you for coming listening and we definitely have time for questions if you've got any so yeah go ahead and use the mics if you have any questions for David line them up up um this is going to roll right into the lunchtime portfolio review so anybody who wants to ask questions for David we want to prioritize those questions and then as those taper off we're going to go into the portfolio review for anybody who is leaving and coming back for launch or or trying to drop in early what we're going to do is we're going to go through anybody who has portfolios that they want to get critiques on and if those don't take us all the way through to the beginning of the talk uh Brennan's talk at what time do we do your talk 145 sorry 145 14545 okay so the next talk begins at45

we'll do portfolio for you till then uh if we run out of portfolios we'll do an open Q&A with anybody who wants to talk to the speakers until that point hello hi so uh I'm kind of in the same boat where I started out like with programming knowledge and then I'm start of going towards the design so I understood some of your constraints um one thing that you mentioned that I haven't explored and I was curious about was you said you had problems with the verticality like you had some things and I was wondering if could expand on some of the experiences you had trying to implement more verticality and what the negatives of that were sure yeah the question is about verticality and the levels and if I could expand on you know some of the experiences I had with that building eldrid uh one of the things I I just kind of skipped over in the slide let me see if I can get back there the when I was talking about like the safe regions in levels the image on the right here is actually the safe region for vertical traversal it had to be wider than the one for for lateral traversal just because like what it takes to move vertically

you know move move up specifically like falling is really easy in games but to move vertically you know Elder actually has a pretty sophisticated movement Suite like you can mantle up onto onto two meter high objects you have powers that can just boost you into the air and things uh a rope gun so you can just shoot ropes into the ceiling and climb up them it has a lot of ways to move vertically but I couldn't count on the player to have any of those at any given time so I always had you know when I would make a module that's supposed to connect upward I would have to build like kind of these really ugly elaborate staircases up and it ended up being kind of like the all the worst rooms in the games are the in the game are the ones where you're trying to move up because there's no space left for anything interesting to to to happen there once I've got that in if I could have done something where like I If I Could Have guaranteed that the player never had to move up and you could just sorry and you just fell between levels then it would actually be a lot easier to do that because you know like I said falling is really easy thank you there um thank you very much it was really interesting thank now um I was

just curious how long I see that you built a lot of tools and do did a lot of tweaking I I just want to know how long did it took you from taking it to the like the first part until that you felt that your tools were already generating like really good levels right so the question is how long did it take I I guess between like kind of the first pass version of this and when it was like when when the levels were good or when the tools were good which are you asking when the levels were good when where you felt like okay I'm not touching it anymore it's good yeah yeah um I'm not sure I can really say like the the development of the game was so accelerated and I was kind of even up until the very end I was still changing things um the I also built the levels chronologically which is kind of not a best practice so like the first world in the game was the first one that I built and I think you can maybe see that in the game of like the the first world is a lot more experimentation like by the end of the game I had kind of figured some of that stuff out but the the tools

actually didn't change too much it was just kind of like I said on that last slide it was figuring out what worked to make a good level and how how to guide the player in a random space thank you hello thank you very much hi I would like to hear any comments or tips for non-scripted people uh in and the downfalls of using or scripting the the procedural system rather than using a dedicated software for generating that thank you was that a question okay uh say for example Houdini who is a dedicated software for procedural generation levels what are the downfalls of actually coding your procedural stuff than using a 3D package right right right I'm sorry I misunderstood what's the advantage of using an existing solution for for procedural systems instead of writing my own so I actually haven't evaluated like I mean I I didn't even know that sort of thing existed I think in in the case of

eldrich it's kind of the same problem I have coming from AI programming of like so much of procedural level generation is kind of innate to what the game is I don't know if there's a solution out there that would suffice for exactly what I was trying to make and also I mean honestly this stuff like I probably spent about two weeks working on the tech and two weeks building levels like a relatively small portion of the time I spent making this game was actually spent on this stuff it didn't take that long to get you know to where to where I was able to start generating levels on this you know if I were trying to do procedural levels on a big budget game and there were much bigger requirements I probably would consider using something like that but I don't know I haven't tried that all right hello um thanks for giving your talk uh sort of more just a overall Development question uh you said you built your own game engine for this game uh how how difficult was that and do you think it was worth it do you think that it gave you tools you needed that you would have had wouldn't have had otherwise or you think you might

have saved time using an existing game engine so the question is uh was it worth building my own engine for this game so I I sort of misspoke I didn't build the engine during the eight months I was building eldrich like that that would have been impossible the uh I I've just kind of been as a hobby building an engine since eight or nine years ago and so it's something that at this point is pretty mature and something that I'm like I work very efficiently in it's kind of like if I had my own licensed engine like that's just how comfortable I am in it and and how capable I am in it so I I wouldn't necessarily recommend it to everyone for me it was kind of like you know I I'm more efficient using my own code than I would be like if I were using something like Unity or unreal but I wouldn't recommend it probably to anyone else I don't know if you have enough experience with unity or unreal but if you want to do something like this in an existing engine do you know would you have any recommendations or good question

um I think there are tools on the asset store or yeah on the asset store for Unity um I I know that some people have done procedural levels like in even going back to like the original Unreal Engine it's sort of different because it tends to make assumptions about like how AI paths are built and things like it assumes that the world is static and not pieced together at runtime um but yeah I'm sorry I can't really speak to existing Solutions it's okay thank you yeah thanks uh if you were going to make a sequel to eldrich how would you modify your algorithm going forward knowing what you now know having made eldrich good question that may be something I'm actually thinking about um so one of the things I wanted to do on eldrich and I kind of just got rid of it because of you know trying to do everything as simple as possible but I wanted to have modules that were different sizes one of the the key constraints here is that every module is just 12 M by 12 M that's not a lot of room to work with and I had planned to have like support where the you know if if I could search the maze

for a fit and I could say like well here's a space where I could put in like a whole big mind shaft and then I could design that as like a sort of unique feature that would be cool in that it would give me a lot more visual variety and sort of cohesion to the levels and if I weren't doing the lovecrafting thing I think that's something I would almost definitely have to do it was just kind of a trade-off I had to make on this game to get it done fast but yeah if I were doing that in the future I think I would want to have the ability to make modules that were different sizes that could still join you know it would still beond the the same grid as the maze I think because that's you know that's just a nice feature to have but yeah the ability to do to do modules that were different sizes would be a big one thanks thanks thank you thank you for the talk uh I was wondering about the overall pacing in your game uh how were you able to shape that did you do play tests or yeah so the question is about pacing in the game and and and how was I able to get into a shippable state I did do a lot of play tests over the course of

development it was also a game where one of my key values was like you know I love immersive Sims like DSX Dishonored the sort of things I mentioned at first and so I wanted to have this sort of value of like play it your own way you can kind of build up any set of powers and weapons and always have a solution to the map which meant I I couldn't really control Pac scene at all I just kind of had to throw a bunch of features in and hope that the experience was good it ended up being kind of a system soup it doesn't have great pacing the only pacing I could do was like you know like I said ramp up the number of enemies over the course of the game you know in between levels I could say and I I could give like you know scheduled item drops and say like in the first world of the game you're not going to get the most sophisticated Powers you'll kind of you'll get there eventually um but yeah like within the level I really had no control for like encounter pacing it was just kind of like stuff's going to happen and I hope it's cool thank cool thank you I think that's everything all right thank you for your time