You know, i've seen a lot of level design issues that can be solved by random terrain/level generation. Such as the size of the gaming world itself. So many development hours are spent building the gaming world, one has to wonder what would happen if you just used a randomized generator to do most of the difficult/time consuming work.

   If you used a random generator to generate a game world, you could more or less make it any size required, although probably not all at once. I got this idea from an older program created back in 2001/2002 for the game Morrowind.

   Then recently I've seen some of Lionhead's designer diaries, and they interestingly enough reminded me of other industry professionals (in various places and media) who also made statement that had to do with the limitations of the game development team. Specifically, creating massive game worlds takes a lot of level designers, which raises development costs. I think the one that caught my attention might have been Peter Molyneux, when he said something about how he kept wanting to include more and more stuff in the game, but was only able to include so much just do to the practicalities and limitations of the development team. Sacrifices had to be made.

   But what happens, if you just used a simple solution, that would allow you to overcome the limitations of the level builders? What happens if you had the ability to create a huge game world? You see what I'm saying, with the right attention to detail the game would certainly be better for it. And it'd set a new standard in the industry. No more tiny game worlds for me to wander around in.

   So that got me thinking about that program I mentioned, and how if a game studio used these random generators to do the bulk of the work, then they could do two things:

A.) have fewer level designers

&

B.) those fewer level designers could then concentrate on key areas.

   Let me explain; if the level builders/desinger were free'd up, then they'd be able to focus on specific areas. When a level builder wastes their time making a grassy field for instance, thats a real waste of their talents. Talents that could be beter applied by making key game areas (cities and towns, mission/quest locations ect) more razzle dazzle impressive. Better attention to detail certainly. It would strictly speaking, allow a development team to create a huge gameworld with far less costs involved.

   In today's market, game studio's are really being pushed to go bigger and better with games. Bigger game enviroment areas (whether they be interior or exterior in nature), because players are constantly exploring the game world, and also better graphics and more features. Except the problem is, that huge game world's take sh*t loads of design time to build, and can really cost money to do.

 

The solution to this problem? Well duh, it's the random generator's I've been talking about.

   I've had some experience with these type of programs and think that without enough work the things could probably create most the game world automatically. You see, it all has to do with the complexity of the generator itself. The more complex and sophisticated it is, the more realistic areas it will be able to develop.

   For instance, if you want to make a forest. Then you need a feature that allows you to designate a certain percentage of the area to be covered by tree's. You'd also need an option to decide on the density of the trees. How many tree's per square acre for instance. Then include some options so you can decide what type of trees and plants are included in the area, and you would have a very basic forest.

   In this way, you could probably create a forest the size of...well, however big you want it to be. This adds realism, because suddenly, the game world isn't restricted to a small place. I remember, even when playing games like Oblivion (The Elderscrolls IV), I remember feeling vagely claustrophobic. And bored as well, since aside from how impressively large the game world is, it's just that big in actuality. Especially with the ability fast travel anywhere on the map.

 

   I remember playing in the Oblivion game world, and thinking that such small areas were unrealistic and not very interesting as a whole. The Great Forest in that game for instance, actually wasn't very big. You could walk from one side to the other in only a few minutes. Whats that mean for me? It means I got bored with that area very quickly, thus the replayability of the game as a whole was diminished.

   Now I know what some game designers are thinking. They probably realize that the output files would require a lot of harddrive space (the GenMod program discussed below certainly did). Well, that might be a problem for consoles, but for pc's it isn't really an issue. Computers these days are practically coming with 100-300 GB of HD space as a standard. So what's the big deal about having a game world that takes up 10 GB? How about 20? It's really a non-issue for PC's. And most gamers would be perfectly willing to uninstall other stuff to make room for a game with that big of a game world.

   So the actual limitation in implemnting the feature, would be the time it took for programmers to create the generator itself. So it's not the cure for all development team woes. Just an improvement.

 

   I've seen a very basic level editor that was developed for The Elderscrolls III: Morrowind, it was made by a game user/fan so it wasn't all that complex, but it could create dungeons of near infinite size. Have you ever had to hack and slash through a 500 level dungeon? I have, and while it was very simply designed, it worked quite well. Took me days to get to the last level (the 500th floor/level whatever you want to call it).

   Is a 500 levels dungeon realistic? No, it's not. But then again it's neither is magic, or any other number of things you see in games. Or if you didn't want to make such a expansive dungeon, just make more of them. Have 500 1 to 3 level dungeons instead. When I say level, I'm refering to the seperated areas of that dungeon.

   That program was called GenMod, and was fairly simplistic as I've already mentioned. The individual who made it also did some work on a version of GenMod that could create exterior areas too (rather than strictly interior areas), however it was never completed. Most likely due to the time involved. However with that said, it's obvious with the right design team it could definitely be done. And manage to look in the end.

   The program I'm talking about was designed to use the construction set, and create output files that could be loaded by the construction set that came with the game. (TES construction set).

   So in conclusion, I'd say that a generator of this scope and complexity is definitely within the realm of possible. There's no doubt it could be done. And I think it could be implemented in a reasonable amount of time (months). Also, costs might not be offset by needing fewer level designers, since you need programmers (more programmers?) in order to build the program. But, I think it'd be both possible and practical.

 

   Now if only the game industry would take a note from me, and actually utilize this concept. We'd raise the bar (standard) all across the industry.

   No more tiny game worlds! At least not in the open-ended RPG's. I have a lot of solutions like this that I'll discuss to some degree or another here on the forums (and in other places on the net).

 

   I'd love to get some feedback from other forum members. What do you all think of the concept?

---

"Whenever cannibals are starving, Heaven, in its infinite
wisdom, sends them a fat missionary"
-Oscar Wilde

"Perhaps it's impossible to wear an identity without becoming what you pretend to be."

-Page 231, Ender's Game

It's interesting technology, but it has a fundamental unsolved problem: Everything that's algorithmically created like this is generic and bland. Oblivion's problem wasn't that the world was small, it was that the world was boring. Pretty, but boring.

Artists and level editors are the only way to make areas that are really awesome, that work well and don't feel, well, randomly generated.

I've seen a very basic level editor that was developed for The Elderscrolls III: Morrowind, it was made by a game user/fan so it wasn't all that complex, but it could create dungeons of near infinite size. Have you ever had to hack and slash through a 500 level dungeon? I have, and while it was very simply designed, it worked quite well. Took me days to get to the last level (the 500th floor/level whatever you want to call it).

That sounds incredibly monotonous and boring to me. I cannot imagine enjoying spending days in the same kind of area, from a level design perspective. Maybe I'm a lot more impatient than I am in that respect, but I don't think I'm unreasonably so.

Is a 500 levels dungeon realistic? No, it's not. But then again it's neither is magic, or any other number of things you see in games.

Realism isn't really the issue. Magic is not realistic, but there is suspension of disbelief to allow you to still enjoy it. 500-level dungeons violate suspension of disbelief.


Randomly generated areas always feel randomly generated. I don't just think that's because of the currently quite bland technology. I think it's because what makes an area interesting is the things that happen there. An area needs to have story to be great. This is why Oblivion fails. Random cave/ruins/mine number 42 has no story, it's ... just another place.


P. S. Nicely started thread.

P. P. S. Ender's Game

I agree. :-)

However, the world around us is randomly generated (in nature anyways). And the world generated is only as and or realistic as the algorithim's that are coded for doing the job.

I agree about the 500 level dungeon as well. But it was a challenge. Which is why I did it. It's also an extreme created by a overly-enthusiastic modder in the fanbase community. However, it's really all a matter of to what extent you want to do something.

Also, a 500 level dungeon might not be quite as unrealistic as you think. (Or as unrealistic I originally mentioned). I just now realized, that some of the catacombs in France and Italy are very very large. I mean, aren't the french catacombs almost as large as the city of Paris? Refered to as the underground?

Also, I'm curious about what you said in your post. You said what makes an area interesting is the story behind it? Well, thats something that wouldn't be a problem. If you think about it, the human level builders would have more time to focus on implementing quests and such if they had something to get a lot of the tedious work out of the way.

Plus, every location can't have a story. It's not realistic either, now that I think of it. Personally, I loved exploring Oblivion, as long as I was able to hack and slash my way around and find treasure and stuff. Though you are right about the oblivion enviroments being boring as in bland (in a not a lot to do sort of way). However, thats not an problem without a means to fix it. With a little hard work and effort, and some ingenuity I think a development team could smooth out issues of that nature.

---

"Whenever cannibals are starving, Heaven, in its infinite
wisdom, sends them a fat missionary"
-Oscar Wilde

"Perhaps it's impossible to wear an identity without becoming what you pretend to be."

-Page 231, Ender's Game

Veni_Vidi_Vici:

However, the world around us is randomly generated (in nature anyways). And the world generated is only as and or realistic as the algorithim's that are coded for doing the job.

Well, not really.

The world around us wasn't randomly 'generated'. It took billions of years to look like it does today, through a complex process no current technology could even begin to conceivably replicate. It was randomly 'evolved' perhaps, or maybe I'm just bickering over semantics. The world around us gives the impression of design through natural processes that shaped the world, such as erosion. I wonder, do we know the algorithm's that govern that?
I could go off on tangents about the concept of beauty, but I shan't.
The point is, to get anywhere close to something looking like our world without the hand of an artist is just...impossible. Like you said "...the world generated is only as and or realistic as the algorithm's that are coded for doing the job." They'd need to be bloody .

Using random terrain generators for some areas is a idea, and I'd be surprised if most games don't use them to some degree. But technology doesn't understand beauty and the best game world designs will always be created by professional artists.

I find myself wondering. Would I rather travel a vast expanse of nothing or remain in a box full of colour and vibrancy?

The latter, most certainly.

---

The only thing worse than beating a dead horse is betting on one.

Delirious wrote:

"Well, not really.

The world around us wasn't randomly 'generated'. It took billions of years to look like it does today, through a complex process no current technology could even begin to conceivably replicate. It was randomly 'evolved' perhaps, or maybe I'm just bickering over semantics. The world around us gives the impression of design through natural processes that shaped the world, such as erosion. I wonder, do we know the algorithm's that govern that?
I could go off on tangents about the concept of beauty, but I shan't.
The point is, to get anywhere close to something looking like our world without the hand of an artist is just...impossible. Like you said "...the world generated is only as and or realistic as the algorithm's that are coded for doing the job." They'd need to be bloody .

Using random terrain generators for some areas is a idea, and I'd be surprised if most games don't use them to some degree. But technology doesn't understand beauty and the best game world designs will always be created by professional artists.

I find myself wondering. Would I rather travel a vast expanse of nothing or remain in a box full of colour and vibrancy?

The latter, most certainly."

Not true on a lot of points. Let's talk about random generation as it has existed so far, and exists today.

The first example I use is The Elderscrolls II: Daggerfall:

I quote,

"Daggerfall is the largest Elder Scrolls game to date, featuring a game world estimated as being 161,600 square kilometres (40,400,000 acres) with over 15,000 towns, cities, villages, and dungeons for the player's character to explore. According to Todd Howard, Elder Scrolls programmer, the game's sequel, The Elder Scrolls III: Morrowind is 0.01% the size of Daggerfall, but it should be noted most of Daggerfall's terrain was randomly generated. Vvardenfell, the explorable part of the Morrowind in the third game has 6 square miles. The Elder Scrolls IV: Oblivion has approximately 16 square miles to explore. In Daggerfall, there are 750,000+ non-player characters (NPCs) for the player to interact with, compared to the count of around 1000 NPCs found in Morrowind and Oblivion. It should be pointed out that the geography and the characters in these later games are much more detailed."

I'll point out several key things:

• It did point out that the game world didn't have as much detail as the later games (Morrowind and Oblivion) which are both strictly human designed and built. But then again, consider several facts:
  • Daggerfall was released in 1996, though it was a little outdated most likely due to budget constraints. So it had graphics comparable to games from 1993 and 1994. Considering this fact, the worlds were actually quite detailed considering.
  • Compared to games of 1993 and 1994 such as Wolfenstein 3D, and Doom...Daggerfall was pretty impressive all across the board.
  • And finally, the game WAS released in 1996. You aren't going to get a lot of detail because of that reason alone.With today's technology, you'd be able to do FAR more.

Other games have also used random terrain generation sucessfully such as:

1. X-COM: UFO DEFENSE: Used random terrain generation to create the landing/battle sites and did it quite well. Given the complexity of the graphics, the enviroments were believable and realistic and also very functional.
2. Sim City Series (including Sim City 4): Can create random maps. They do looks pretty realistic.

I also note some more recent advances in random terrain generator which can be applied to video games.

Such as, www.cs.ubc.ca/~van/GI2005/Posters/Wijaya_Callele_GI2005_Poster_Abstract.pdf

 

I've gone ahead and copy and pasted the text here so you don't have to download that file:

"Semi-Random Terrain Generation

Mariatta Wijaya

&

David Callele

 

Department of Computer Science

University of Saskatchewan

 

Abstract

We present a voxel based system for semi-random terrain generation. Cellular automata rules guide growth, erosion, and transformation phases. Emergent behavior of the cellular automata gives the appearance of randomness at the micro scale while producing repeatable results.

Key words: Voxel, terrain generation, cellular automata

1 Introduction
Asset creation is rapidly becoming the most costly element in video game production. Player expectations for virtual worlds, commonly referred to as levels, are constantly growing. The return on investment is questionable: portions of levels, sometime even entire levels, are never seen by many players yet if they are not included, the game is likely to receive weaker reviews.  This work presents a system for semi-random terrain generation. At the macro level, the terrain is strongly patterned – in this case, the terrain generation process is initialized with a 2D maze construct. At the micro level, the terrain presents the player with the appearance of natural randomness. Ideally, the player would be unable to detect the maze within the final terrain.

2 Related Work
While there are many possible approaches to terrain generation, we seriously considered only mesh and voxel based approaches. Mesh techniques, such as subdivision surfaces, multi-resolution meshes, and adaptive tessellation, typically deform an initial planar mesh to form the terrain. While we could achieve our goals using these techniques, the numeric complexity of these approaches (particularly with respect to managing texture coordinates for scenarios such as caves or tunnels) precluded them from further consideration.

The second approach is voxel based [2], [1], [3]. A voxel approach has the benefit of being a discrete approximation, the associated mathematics are in N. Texture coordinates are always locally scoped – with proper planning, all textures will tile independent of orientation.  Unfortunately, a voxel approach generates significantly larger data sets than a mesh based approach and visualization can be challenging for large data sets.

In production, a combination of techniques is in order.  The user can generate a discrete approximation to the terrain with our voxel based system. Then, as a final optimization phase, a continuous surface mesh could be generated from the discrete approximation [4].

3 Requirements
The system must automatically generate a discrete approximation to outdoor terrain for use in a real-time video game. We must be able to control the macro level features of the terrain. At the micro level, the system should generate terrain that appears similar to the natural variation found in the physical world. The terrain generation process must be repeatable – for a given set of inputs, we must generate the same sequence of outputs. True randomness is not acceptable for we must be able to analyze, play test, and verify the suitability of the resulting terrain.  Finally, the terrain should be interesting to a player, the player should not be able to detect any patterns in the results.

4 The Terrain Generator
Our solution is a three pass terrain generator that applies Cellular Automata (CA) rules to create the terrain out of fundamental building blocks (Figure 1(e)). Each building block is bounded by the volume of a unit cube and can be in one of four possible orientations.

The system is seeded with an initial maze layout (Figure 1(a,c)). As the terrain is generated, the walls of the maze are transformed into hills and mountains. During the GROWTH phase, the walls of the maze are created by adding cubes beside existing cubes and stacking cubes on top of existing cubes. The resulting structures are strongly pyramidal in nature. During the EROSION phase, cubes are removed from the terrain to provide greater variation (Figure 1(g)). During the TRANSFORMATION phase, the type and orientation of the cubes on the surface of the terrain are changed to more closely approximate natural randomness.

The current system has approximately 100 CA rules.  The application of each rule is controlled via an associated probability attribute. The rules are stored in an XML format to promote experiential investigation.  CA are well-suited to our requirements. The macro scale attributes of the terrain are readily achieved by simple growth automata. The desired micro scale attributes appear as a byproduct of the emergent behavior exhibited by the interaction of so many cellular automata rules.  The emergent behavior comes at the price of reducing fine control over the resulting surface. However, if we assume that the output of this system is only a discrete approximation to the final terrain, any micro control deemed necessary can be achieved in subsequent operations.

5 Results
Typical results are shown in Figure 1. Figure 1(a) is the seed maze for a low resolution terrain. Figure 1(b) is a top view of the resulting CA output and Figure 1(f) is a 3D visualization of the same data. Figure 1(c,d) are the seed maze and top view of a high resolution terrain.  Figure 1(g) shows the low resolution terrain at the end of the erosion phase. Figure 1(h) shows the final result of the high resolution terrain generation of Figure 1(d).

6 Conclusions and Future Work
We have demonstrated the feasibility of generating semirandom terrain using voxel techniques controlled by cellular automata rules. The resulting terrain is a discrete approximation, suitable for conversion to a continuous surface or for use as inspiration to level designers and level modelers.

Since the CA rules are stored in XML, modifying them is easy which promotes experimentation and rapid prototyping.  In the future, we want to add more building blocks and their associated rules – the next phase will have a dozen geometric primitives and support for trees.  Finally, we will add mapping the discreet terrain to a continuous surface and strongly consider integration into a game engine.

Acknowledgements
Our thanks to nVIDIA for their support of our research.

References
[1] Stephane Gobron and Norishige Chiba. 3D Surface

Cellular Automata and Its Applications. The Journal

of Visualization and Computer Animation, 10:143–

158, 1999.

[2] N. Greene. Voxel space automata: modeling with

stochastic growth processes in voxel space. In SIGGRAPH

’89: Proceedings of the 16th annual conference

on Computer graphics and interactive techniques,

pages 175–184, New York, NY, USA, 1989.

ACM Press.

[3] Sidney W. Wang and Arie E. Kaufman. Volume
sculpting. In SI3D ’95: Proceedings of the 1995 symposium

on Interactive 3D graphics, pages 151–157,

New York, NY, USA, 1995. ACM Press.

[4] John R. Wright and Julia C. L. Hsieh. A voxel-based,

forward projection algorithm for rendering surface

and volumetric data. In VIS ’92: Proceedings of the

3rd conference on Visualization ’92, pages 340–348,

Los Alamitos, CA, USA, 1992. IEEE Computer Society Press."

 

 

Now that was written in 2005. So you can see that more advanced methods are coming out. And I was actually quite impressed with that process that David Callele described and I think it would work.

Finally in conclusion, I'm going to bring up Daggerfall again. You know, the game was ahead of it's time, and had 700,000 NPC's. 40 million square acres of game world? That's beyond impressive. And a note on that is that the game got fairly reviews despite it's limitations.

---

"Whenever cannibals are starving, Heaven, in its infinite
wisdom, sends them a fat missionary"
-Oscar Wilde

"Perhaps it's impossible to wear an identity without becoming what you pretend to be."

-Page 231, Ender's Game

Additionally, I've found another interesting development.

http://www.cajomi.de/GeoControl/geocontrol.htm

This is in fact a random terrain generator of immense capacity. And they claim it's been used in EA's Battlefield: Bad Company. Which has already been released as of a month or two ago.

If you don't think thats impressive, check out their new Isolines demonstration!

http://www.cajomi.de/GeoControl/isolines1.html

The future is already here.

:-)

I've been using these things and I'm getting pretty at them.

I think this proves my point about it being possible.

Also of interest is Terragen.

http://www.planetside.co.uk/terragen/

---

"Whenever cannibals are starving, Heaven, in its infinite
wisdom, sends them a fat missionary"
-Oscar Wilde

"Perhaps it's impossible to wear an identity without becoming what you pretend to be."

-Page 231, Ender's Game

 

Notably, these are images of some of my work from terragen.

They are rendered from completely randomly generated terrain maps. Designed to be a canyon out in a hardpan desert. Very basic color palette involved.

---

"Whenever cannibals are starving, Heaven, in its infinite
wisdom, sends them a fat missionary"
-Oscar Wilde

"Perhaps it's impossible to wear an identity without becoming what you pretend to be."

-Page 231, Ender's Game

By "Random" I'm guessing you actually mean "Procedural"... ?

---

CTG ... Blog

That's a very interesting read. I have oblivion and have explored it extensively, and agree with you that it is quite bland, but only up close. It's similar to what Peter Molyneux said in the "The Room" presentation: game design is moving towards the grandiose - huge expansive areas visible from above. That is one of the things I found awe-inspiring about Oblivion - heading to the Jerall Mountains and looking down to see the rest of the land spread out before me.
What a random level/terrain generator needs is to include the small details that The Room focused on - high definition textures and many more different permutations of the random environment. No-one likes seeing a uniform stone wall that repeats every metre, and neither do people enjoy playing a game when they see the same repeated building over and over - it destroys the temporary disbelief in reality and makes us realise the game is fantasy.
Also drawing attention to the story element, in the new gameplay trailer for Diablo 3 I believe that random events were mentioned - I'm not sure but I think the part where there was a ring of dark cultists that summon demons was one. It looked very integrated with the environment and if it is as complex as it looks at first, then this could be amazing.
Diablo 2 also had randomly generated wildernesses - granted there were no height differences in the terrains but it all flowed very well with the texturing of the ground and locations of enemy camps etc. If they applied the same texturing to a randomly generated height map, I don't think the result would be too bad.
I'm no advanced programmer, but for an extremely simple height map capable of, say, ME, could a flat terrain have random vertices changed and then smooth all the ones around a random amount to create random bumpy or smooth hills and dips? I know this in no way would look professional, but I was just wondering if that was possible.

paulecoyote:

By "Random" I'm guessing you actually mean "Procedural"... ?

lol.

quite right.

i suppose they aren't completely random.

;-)

Something I hadn't mentioned earlier is that Geocontrol can create complicated height maps that can then be exported to Terragen to give more specific results. Because Geocontrol is advancing in capability (Isolines anybody?) I think procdural might be a nice term to describe the whole process.

---

"Whenever cannibals are starving, Heaven, in its infinite
wisdom, sends them a fat missionary"
-Oscar Wilde

"Perhaps it's impossible to wear an identity without becoming what you pretend to be."

-Page 231, Ender's Game

dawesbr:

That's a very interesting read. I have oblivion and have explored it extensively, and agree with you that it is quite bland, but only up close. It's similar to what Peter Molyneux said in the "The Room" presentation: game design is moving towards the grandiose - huge expansive areas visible from above. That is one of the things I found awe-inspiring about Oblivion - heading to the Jerall Mountains and looking down to see the rest of the land spread out before me.
What a random level/terrain generator needs is to include the small details that The Room focused on - high definition textures and many more different permutations of the random environment. No-one likes seeing a uniform stone wall that repeats every metre, and neither do people enjoy playing a game when they see the same repeated building over and over - it destroys the temporary disbelief in reality and makes us realise the game is fantasy.
Also drawing attention to the story element, in the new gameplay trailer for Diablo 3 I believe that random events were mentioned - I'm not sure but I think the part where there was a ring of dark cultists that summon demons was one. It looked very integrated with the environment and if it is as complex as it looks at first, then this could be amazing.
Diablo 2 also had randomly generated wildernesses - granted there were no height differences in the terrains but it all flowed very well with the texturing of the ground and locations of enemy camps etc. If they applied the same texturing to a randomly generated height map, I don't think the result would be too bad.
I'm no advanced programmer, but for an extremely simple height map capable of, say, ME, could a flat terrain have random vertices changed and then smooth all the ones around a random amount to create random bumpy or smooth hills and dips? I know this in no way would look professional, but I was just wondering if that was possible.

Terragen can produce realistic/professional landscapes randomly. Or perhaps procedural if you prefer. lol. And Terragen and Geocontrol both produce heightmaps, complicated ones at that. And both programs can apply texture/terrain details.

So to answer your question, yes.

Terragen 2 is out, and it's even more complex than Terragen. I'm getting to be fairly skilled with it, but it's one of those programs that could take a lifetime to master.

---

"Whenever cannibals are starving, Heaven, in its infinite
wisdom, sends them a fat missionary"
-Oscar Wilde

"Perhaps it's impossible to wear an identity without becoming what you pretend to be."

-Page 231, Ender's Game

I fail to see why procedural should equal boring.

It's a matter of layers. Landscapes can perfectly be procedurally generated. You can develop algorythms as complex as you want to create realistic, even dramatic landscapes.

Heightmap - Water - Vegetation - Animals, all can be automatically generated. By the way, in my opinion they did a pretty job in Oblivion.

After that it gets more complicated. Cities, roads, caverns, outposts, ruins. Their locations, layout can all depend on the (hi)story of the game. So you'd need some way of combining the storyline with your world-generator.

Imagine implementing a timeline, where you describe the history of your game-world from the very beginning, maybe even before cities where founded. This timeline can be used by your world-generator to create a very realistic world, and by adding more and more detail to the timeline, you will add more and more "life" to your procedural virtual world.

This is ofcourse not at all trivial to implement and would be a project on itself, which is probably why no one has tried it before. But I don't believe that it's impossible, and with increasing capabilities of hardware, I think one day we'll reach a threshold that this kind of high-level game design becomes a neccesity.

It's not unlikely that we're already at that point. Spore is about to come out(at last) and already implements some of the ideas mentioned in this thread.

---

Hapiness isn't hapiness without a violin playing goat

Remember Diablo 2? The levels there were randomly generated... It looks on paper, every play-through in a different world etc., but the experience was the same.

When PM said that they had too many things they wanted to put in compared to their capacity, I don't think it was strictly a "space" problem in the game. Rather I think it has to do with the amount of unique material they would like to be in the game. I agree that you can make areas work well from a procedural perspective, perhaps even quests, but it leaves no room for a developer to surprise you or create something innovative.

I think there are two kinds of games: Those that play like games and those that play like movies. Unreal Tournament vs. Max Payne. Diablo 2 versus Morrowind. One type is trying to tell a story, the other is keeping you amused with the game mechanics. For one, endless but generic material is a plus, for the other, it's death.

You could always do like world of warcraft or something similar and mix the techniques. What I mean is, assuming you coded up a design studio for terrain, etc, based on height fields and texture splatting, you setup a generator inside the editor which will give you a foundation based on heuristics and rules to invent a "Believable" terrain, then you massage what it generates to be custom. It would definitely save time on a designer and give a uniformity to the terrain, but also be able to be modifier to have the unique feeling you want from an area by a designer.

I've seen an awful lot of research work thats been done on procedurally generated game worlds (we've even done some ourselves at LH). Ultimately the drawbacks still tend to outweigh the benefits. The main drawback, as has been cited already, is that it lacks the 'human touch'.

If the goal was world building alone, then there's nothing wrong with procedurally generating everything. But for a videogame it's not. The goal is to create a playspace that serves the game design and is 'fun'. Random very rarely equals fun.

Even Diablo, which procedurally generates most it's levels, has a lot of sections that are pre-built and then dropped in in their entirity.

The real world might be procedural (in a way) but the real world doesn't exist purely for the entertainment of the people in it.

One area in games that does see an awful lot of procedural work though is texturing...

The Infinity engine uses procedurally made terrain, but being a space-sim the player rarely interacts with terrain features on any particularly deep level (at least, I don't think it's planned to - it hasn't been released yet; one of the things that appeals to me is that it's a pet-project, not a commercial one and so has no deadlines or anything - it'll be released when it's finished). Like has been said before, in order to put features in that allow a particular story line to be played out, one cannot do without the human touch. On the other hand, you could generate a proceedural terrain and make a story to fit it perhaps?

I would imagine a proceedural vegetation-maker would be a pretty idea mind, like the one used in Vue as this would save a lot of time, especially in non-fps games where the player might have a much wider/far-reaching view.

Actually talking of vegetation: I know It'd likely take up vast amounts of either disc space or setup-time, but I wish that trees (and all vegetative objects for that matter) in games were generated proceedurally. Much as I think BW2 is an amazing looking game, I do wish there was variety within in the tree types.

---

Large Hadron Collider - because fleshlites are for wusses

I've written terrain generators using various algorithms. I use those tools to give me a place to start, but then I manually modify the landscape a little more and add the "interesting stuff" after that. I tried automating the placement of the "interesting stuff", but I always find that I want to tweak it anyway.

---

http://thnogg.mybrute.com