Procedural generation, and why it's better than hand crafted assets.

[lysan]

Ok, you clicked and now you're here, thank you for joying us in this adventure, down the procedural rabbit hole.


The most common answer to procedural generation are, it's faster, you get more, however it's boring and copy paste. Hand crafted is a slow process but at the end, the result will be more interesting and realistic.

I need to step in here, that is just not correct, we have seen in this very forum many examples of interesting PG. The so called problem are rooted some where else. First we need to be sure we talk about the same thing. PG has been use in many applications, and we all know PG from older games. Some were good, other not so good.

Lets go down into the rabbit hole.

The entrance

On the fly GPU generation, this is not a new technique, this has been around for some time now. As our CPU and more important GPU power increase, we will be able to make more complicated models to generate.

This project was from 2002, however there are even earlier projects.

Infinite Fun Space

The arrival

The paper was released in 2007 and the video uploaded in 2010.


[video=youtube;-DBNPlYWGa8]https://www.youtube.com/watch?v=-DBNPlYWGa8#t=26[/video]


Wont you stay for tea?

Physically-based Detail Generation.
I believe that ED use this method, maybe combined with other magic tricks.

Erosion, plate tectonic psychics, well Erosion we can't see right now, so we will need to leave hydrology out for now. For the interested reader, read all about it here.

A small video simulation of the tectonic plates.
[video=youtube;bi4b45tMEPE]https://www.youtube.com/watch?v=bi4b45tMEPE[/video]

"My dear, here we must run as fast as we can, just to stay in place"

Rock's, you can't make rocks. Rocks must be made by hand crafting, because it will not look like a rock if its PG. Well ok, I hereby give you rocks.

Rocktools, yes that tool was made a long time ago, and it can make some very nice rocks.

Spoiler

Rocks add immersion, and they are also interesting from a geo-structural POV.

The Queen

Maybe it's here the head is chopped off, however before that, we need to read the massive paper, including fascinating math regarding PG and how much work that actually is needed to do it right. My point is that if done right, the outcome will be as outstanding as mother nature or pretty close.

I give you the,

Interactive Terrain Rendering: Towards Realism with Procedural Models and Graphics Hardware

Thank you for your attention, wish you a great day.

 

[Avago Earo]

Very interesting. Thank you.

The last link doesn't work for me though.

 

[Ian Skippy]

Cool post, and I agree. After all, mother nature is PG.

Very interesting. Thank you.

The last link doesn't work for me though.

Its links to his local harddrive.

 

[lysan]

Very interesting. Thank you.

The last link doesn't work for me though.

Corrected!

 

[Derath]

I think eventually PG will trump curated content, but right now they are so far apart its unbelievable.

The thing is a game is not google earth, its about the content and this is where PG falls down - they can make complete and fantastic simulations of the world, because it runs on a certain group of systems. Now have it try to make a dungeon.

It doesn't work, because you can't create a clever dungeon with PG unless you've invented AI, you can have lots of custom created puzzle pieces for the simulation to put together, but that isn't PG its a mashup.

Take a game like starbound, every planet is bananas its completely bonkers how much variation there is, but you repeat the dungeons & general creations much faster than you'd expect and you notice it more because they are arguably the most important piece of content.

Love the terrain synth video btw its amazing how far they've come with this stuff.

 

[Kayweg]

Nothing against PG as a terrain/gameworld creating tool, but that's still just the backdrop, the landscape, the playing field if you will.
Somebody then needs to come along and put an actual game in there, because (imo) that bit PG isn't well suited to do, and i think ED bears some evidence of that.
I also have my concerns that NMS will run into the exact same problem.
We may look back in a few years, and conclude that PG turned out to be a withered branch on the evolutionary tree of game development....while for other applications it became quite useful.

 

[Jaggedmallard]

I think eventually PG will trump curated content, but right now they are so far apart its unbelievable.

The thing is a game is not google earth, its about the content and this is where PG falls down - they can make complete and fantastic simulations of the world, because it runs on a certain group of systems. Now have it try to make a dungeon.

It doesn't work, because you can't create a clever dungeon with PG unless you've invented AI, you can have lots of custom created puzzle pieces for the simulation to put together, but that isn't PG its a mashup.

Take a game like starbound, every planet is bananas its completely bonkers how much variation there is, but you repeat the dungeons & general creations much faster than you'd expect and you notice it more because they are arguably the most important piece of content.

Love the terrain synth video btw its amazing how far they've come with this stuff.

This, PG just isn't advanced enough to create actually interesting worlds, sure you'll be impressed at first but eventually you realise that theres no actual content or interesting stories for things and eventually things start repeating. Sure this star system might have 2 earth likes and 3 stations but theres nothing unique about it until the devs step in and assign rares and a system description or maybe a CG, even things like dorf fort that procedurally generate stories don't do them as well as actual hand crafted ones.

PG is great for terrains and the underlying stuff but not beyond that

 

[Lily from anim.o.v.e]

I think eventually PG will trump curated content, but right now they are so far apart its unbelievable.

The thing is a game is not google earth, its about the content and this is where PG falls down - they can make complete and fantastic simulations of the world, because it runs on a certain group of systems. Now have it try to make a dungeon.

It doesn't work, because you can't create a clever dungeon with PG unless you've invented AI, you can have lots of custom created puzzle pieces for the simulation to put together, but that isn't PG its a mashup.

Take a game like starbound, every planet is bananas its completely bonkers how much variation there is, but you repeat the dungeons & general creations much faster than you'd expect and you notice it more because they are arguably the most important piece of content.

Love the terrain synth video btw its amazing how far they've come with this stuff.

you need both, you need generation doing the main work and then a mudularity preset system for dungeons and POI's that get integrated into the generated landscape followed by some "cleaning" to make them both be smooth into each others.

Play planet explorers, the random generated maps can have some so unqiue awesome things it's amazing.

 

[lysan]

I think eventually PG will trump curated content, but right now they are so far apart its unbelievable.

The thing is a game is not google earth, its about the content and this is where PG falls down - they can make complete and fantastic simulations of the world, because it runs on a certain group of systems. Now have it try to make a dungeon.

It doesn't work, because you can't create a clever dungeon with PG unless you've invented AI, you can have lots of custom created puzzle pieces for the simulation to put together, but that isn't PG its a mashup.

Take a game like starbound, every planet is bananas its completely bonkers how much variation there is, but you repeat the dungeons & general creations much faster than you'd expect and you notice it more because they are arguably the most important piece of content.

Love the terrain synth video btw its amazing how far they've come with this stuff.

Regarding caves, it's actually still being in its infancy, we don't know how far the projects have come.

http://www.resurgentsoftware.com/rosettastal.htm

The above link is a data collecting project for mapping 3D caves and overhangs.

 

[Derath]

you need both, you need generation doing the main work and then a mudularity preset system for dungeons and POI's that get integrated into the generated landscape followed by some "cleaning" to make them both be smooth into each others.

Play planet explorers, the random generated maps can have some so unqiue awesome things it's amazing.

Yeah but PG is used to save time and its excellent at the job, but there isn't anything thats PG that couldn't be created by a good team its just a better management strategy to have a robot create the background and have the humans make the content that goes in the background.

Not really disagreeing with you here I'd certainly use both if i was making any sort of open world game etc, but i'm yet to find a game that has heavy PG that i don't notice repetition extremely early in the content that matters to me, sadly I don't find different heights of geological features interesting though more power to you if you do.

- - - - - Additional Content Posted / Auto Merge - - - - -

Regarding caves, it's actually still being in its infancy, we don't know how far the projects have come.

http://www.resurgentsoftware.com/rosettastal.htm

The above link is a data collecting project for mapping 3D caves and overhangs.

Again the thing is a cave is not a dungeon, Landmark had PG caves with "content" in, it just didn't work and at least a little bit behind why EQNext was cancelled (other than obvious investment group buyout). The thing is the AI has no idea where to put loot for example, or traps, or monsters, it looks like a kid went wild with the content creation kit.

 

[Barking_Mad]

I have a problem with the rocks in ED. They're just a bit unconvincing, in the main.

 

[lysan]

I have a problem with the rocks in ED. They're just a bit unconvincing, in the main.

Well, maybe they should use ROCKTOOL

 

[Ian Skippy]

I think people in generally are too blinded by the present to see the future. There is no theoretical reason why PG cant create quests, interesting puzzles, backgrounds, personalities, or anything we know in life. Way back in the day it was argued 3D engines might be used for maze-games, but there couldn't be real-time NPCs because there would be no way an AI could finds it way through 'the cyberspace' (love the terms from back then). There are countless of these examples. In general I hold the following to be true: if there is no theoretical hard-limit to prevent us from reaching some goal, we'll get there. Its just that the exact route cannot be seen before others have cut away the branches.

 

[Lily from anim.o.v.e]

I think people in generally are too blinded by the present to see the future. There is no theoretical reason why PG cant create quests, interesting puzzles, backgrounds, personalities, or anything we know in life. Way back in the day it was argued 3D engines might be used for maze-games, but there couldn't be real-time NPCs because there would be no way an AI could finds it way through 'the cyberspace' (love the terms from back then). There are countless of these examples. In general I hold the following to be true: if there is no theoretical hard-limit to prevent us from reaching some goal, we'll get there. Its just that the exact route cannot be seen before others have cut away the branches.

theoretical reaosn is that evertyhign added int he future basically destroys all the current effort and love people have put into exploration by simply nullifying what they did.

 

[2plex]

I think people in generally are too blinded by the present to see the future. There is no theoretical reason why PG cant create quests, interesting puzzles, backgrounds, personalities, or anything we know in life. Way back in the day it was argued 3D engines might be used for maze-games, but there couldn't be real-time NPCs because there would be no way an AI could finds it way through 'the cyberspace' (love the terms from back then). There are countless of these examples. In general I hold the following to be true: if there is no theoretical hard-limit to prevent us from reaching some goal, we'll get there. Its just that the exact route cannot be seen before others have cut away the branches.

Because of Kurt Gödel's Incompleteness Theorems.

 

[Wylie28]

I think people in generally are too blinded by the present to see the future. There is no theoretical reason why PG cant create quests, interesting puzzles, backgrounds, personalities, or anything we know in life. Way back in the day it was argued 3D engines might be used for maze-games, but there couldn't be real-time NPCs because there would be no way an AI could finds it way through 'the cyberspace' (love the terms from back then). There are countless of these examples. In general I hold the following to be true: if there is no theoretical hard-limit to prevent us from reaching some goal, we'll get there. Its just that the exact route cannot be seen before others have cut away the branches.

This is why im so exciting for No Mans Sky, ive been following that game since VGX and i have been blown away at how good their generation techniques are. The worlds are very convincing and unique, the creatures arent just copy and pasted models but use something similar to mmo character creation, they even use color theory. I think that once that game comes out many people will start to realize that eventually computers will make content just as good as humans. I always tell people who bash on procedural generation that human devs use procedures too.

 

[Lobstris]

I do wonder why more effort hasn't been put into PG for (at least side-) quests in games - e.g. to generate backstory, overall rational for a quest, plot points, branches if and where necessary. Sure things like Skyrim and FO4 have the Radiant AI system for generating 'one-shot' quests that amount to 'go there, do <that> to character X, come back for reward', but what about mission branching - OK if you spawn two new missions for every win/lose condition from an initial point, things get out of hand quickly, but has anyone played with things like L-systems for story generation, from a system grammar based on linking key story tropes and elements?

In answer to my own question - possibly because it's really hard to do and get consistent, coherent results. (Bet there are a few billion howlers of planets in amongst NMS's quintillions - in the same way that Elite's Stellar Forge throws up the occasional oddity (e.g. how many times is the Roche Limit seemingly ignored without visible consequence?)

 

[Jaggedmallard]

I think people in generally are too blinded by the present to see the future. There is no theoretical reason why PG cant create quests, interesting puzzles, backgrounds, personalities, or anything we know in life. Way back in the day it was argued 3D engines might be used for maze-games, but there couldn't be real-time NPCs because there would be no way an AI could finds it way through 'the cyberspace' (love the terms from back then). There are countless of these examples. In general I hold the following to be true: if there is no theoretical hard-limit to prevent us from reaching some goal, we'll get there. Its just that the exact route cannot be seen before others have cut away the branches.

The issue with PG is when it starts getting really good you start getting dangerously close to proper AI technology especially if its creating completely unique things, stories and dialogue. Theres a Roald Dahl short story about a machine that effectively does what you're thinking of and once it got good enough it effectively rendered artists obsolete as once PG can create quests, background lore and stories on par with hand crafted stuff you no longer need to waste time paying people to handcraft.

 

[lysan]

You will always need the artist to check and tweak the stuff. And we are only talking "nature" generated stuff. Buildings and other man made items will continue to need handcrafted art and assets.

 

[Jmanis]

The most common answer to procedural generation are, it's faster, you get more, however it's boring and copy paste. Hand crafted is a slow process but at the end, the result will be more interesting and realistic.

This isn't really right. We were getting interesting, not-boring stuff out of the procedural generator, we (the playerbase) just screamed for it to be nerfed so some players who tried to play smarter didn't get some advantage over homogeneous grinding. See "Seeking Luxuries" (a result of economic boom generation), Archon-Torval slave runs (a result of political variances in powerplay) and the attempts to nerf Robigo runs (generation of actual concentrations of specific high value mission types due to the procedural seeding).

We need more of this kind of variance around the place, not less, and the procedural generator is capable of it,, we just demanded FD neuter it.

PS for anyone that thinks the Seeking Luxuries nerf was justified, I guess you've never heard of Seeking Goods, Seeking Weapons or any of the other things that provide identical profit margins, and are still in the game.