About the Author
Mass Effect
Final Fantasy X
Batman:Arkham City
Borderlands Series
Weekly Column
Champions Online
World of Warcraft
DM of the Rings
Good Robot
Project Frontier

Building worlds with a single click

By Shamus
on Wednesday Jul 20, 2011
Filed under:


Edge Magazine has a nice article talking about procedural worlds. Project Frontier makes an appearance, as well as Procedural World by Miguel Cepero.

The two projects make a nice contrast, and the fact that they are so different underlines one of my main points: We have not even begun to explore this technology. Ten more programmers could show up. launch projects of their own, and I doubt we’d have a lot of overlap. There are simply so many possibilities and so many approaches to building things. We have long since solved problems like pathing, bump-mapping and in-game physics, to the point where everyone knows how to do those things. But nobody can say definitively, “THIS is the best way to generate topography” or “HERE is the ideal system for making trees”. There is so much ground to cover and so few people working on it. (When compared to the number of people exploring the nuances of “crouching behind brown chest-high wall” technology.)

Comments (106)

  1. Nathon says:

    I read “There is so much ground to cover” as “There is so much ground cover”. It seems like there’s actually a shortage of ground cover, like you said. Clever.

  2. Jjkaybomb says:

    What was really interesting to me was when you said you were confused why graphics were were constantly focused on as technology advanced. With so many possible paths, why graphics indeed… Are they just easier to polish and make better? Programming better AI sounds like a frustrating process… but can it be more frustrating than making shinier graphics?

    • thegrinner says:

      I imagine it’s due to graphics being one of the most visible things in a game. Every review has screenshots, and that’s what catches peoples attention. Sure the review might slam AI or balance, but if it looks pretty it seems like people have a sort of switch saying it won’t be so bad.

      • Daemian Lucifer says:

        Besides,look how fast f.e.a.r. went into obscurity,and that game still has the best ai of all the shooters.

        • Klay F. says:

          And look at how people STILL talk about Crysis (the first one), even though its AI is (mostly) shit.

        • Max says:

          FEAR had pretty good AI, but it had some of the most boring, endless corridor, level design.

          Several years ago, I read a paper on the AI in FEAR that I would link if I can find it again. From what I remember, the AI isn’t really that good, they just make it seem smart, because of the level design.

          For example, most people are pretty impressed by the flanking that the enemy soldiers do. The AI has no knowledge that it is flanking, it thinks its just advancing to the next cover, along preset paths chosen by the level designers. The Level designers are mostly responsible for making the AI seem better than it is.

          I believe the Paper was called “3 states and a plan” so if you can find that using Google, it will have more information and it will be better than my memory.

          The point being, that irrespective of the AI, people(like me) probably got bored with FEAR because of the boring corridor level design. If they had more open level design, I’m not sure the AI would hold up.

      • Adam F says:

        People respond to visuals. Even fruits and vegetables are bred for looks, not the obvious characteristic: taste.

        • Chargone says:

          ehh, actually, fruit and vegies are usually bread for some combination of size and whatever traits are appropriate for how that thing is cooked. try using the wrong sort of potato for a given cooking method and you’ll see what i mean. (some sorts aren’t meant to go in stew. they dissolve completely. others do or don’t roast better, and so on)

          then there’s the rediculous number of different types of apple. same as potatoes, and flavour variation.

          though i’ll grant you there are some random ones done purely for aesthetics as well.

          • Adam says:

            I mean, specifically, that a large part of the breeding of the produce you will find in a supermarket is for looks–second only to breeding for ability to ship. Just compare your average store bought tomato or strawberry to a home-grown variety, which will be smaller, uglier, and tastier.

            Heh, that made me think of indy games as the smaller, uglier, and more-fun cousins of mass market games.

      • MichaelG says:

        I think a lot of it is due to management and marketing. They aren’t playing the game, just looking at it, so they concentrate on what they can see.

        Plus if you are some older manager, you probably *can’t* play a top of the line shooter. But you can criticize the look.

        Marketing knows that visuals sell the games, not more subtle points about game play.

        • krellen says:

          Gameplay sells games to me.

          • vukodlak says:

            Yeah, but you’re not exactly the AAA target audience. MW2 Bro? ;)

            • Chargone says:

              i’m not sure what this stands for, but i Want it to be Mech Warrior 2. clunky old graphics asside, that game was FUN. can’t for the life of me remember what happened to my PC copy though, and the PS1 version had extremely limited (or possibly no) customisation and no multiplayer.
              (strange weapons load outs in multiplayer was crazy fun. that’s right, i kill you with the PPC… IN MY FOOT! :D)

              • Wtrmute says:

                Great guess; I loved MechWarrior. Unfortunately, in this context, “MW2” stands for Call of Duty: Modern Warfare 2, which you might remember by the stunt that publisher Activision pulled on developer Infinity Ward…

    • Alexander The 1st says:

      I think one giant important thing to keep in mind is that with graphics…they’re portable improvements.

      Just try and port Metal Gear Solid’s AI to Call of Duty, and you’ll get…interesting results.

      Whereas if you wanted to, you could mod the Metal Gear Solid graphical engine to work with Call of Duty. No real reason why you’d want to do it, as the graphics would be significantly worse (Unless this was MGS4, but I’m assuming MGS1 here), but if you did it vice versa (Call of Duty graphical engine into Metal Gear Solid 1)…instant performance boost (Assuming you haven’t updated the polygon count/texture filesize to meet the increase in technology, etc.), less graphical glitches (Well, possibly), and maybe reflections, etc. that aren’t given directly by the game data, but come from the engine.

      When doing a new game, it’s the one thing you’re guaranteed to salvage. Consider the Source engine – it can run Left 4 Dead(2), Half Life (2): Episode (2) (As I understand, the Source engine runs all Half Life games?), Alien Swarm, Portal 2, etc. All with very different AI workings (Half Life has NPC AI on top of enemy AI, which is significantly different, and Portal (2), ironically, is probably one of the easier AI systems for them to program.), yet the same core graphical engine, sometimes with updates that showcase a new feature (Portal rendering, for example), but that stuff can ported over.


      There’s a similar issue with procedural generation – Shamus is doing his for a presumably FPS-type game, but with height-maps at its core, it can’t do caves (From what I understand) without a major re-write. Even without height maps, you don’t want flowery grass growing in a dark and dreary cave…Unless you want it to.

      EDIT: Just a thought, but if you wanted developers to take procedural generation seriously, you may have to adapt the Project Frontier into two games – build one out of it with some gameplay, AI, etc., then swap them out into a completely different game (i.e. Halo/Fallout for one game, Red Alert/Civilization the next). Then they could see that it’s not all about re-writing the procedural elements.

      • Fat Tony says:

        yeah what this guy sain in his edit, this if done right could (plausibly) change peoples way of thinking if you could get terrain and AI to run procedurally and use them in a plethora of ways, while still using the same core system that would be astonishing.

        • StranaMente says:

          Well I think the success of minecraft and terrariais quite an example of what you can do with procedural content. And yet we scratched the surface. Surely everybody is looking at minecraft, but hopefully it won’t take long to realize the potential of this idea applied in other game types.

          • RejjeN says:

            I’m not completely sure if Terraria IS procedurally generated. I mean yes, the worlds are generated when you create them but after that point they don’t change or grow on their own…

          • Alexander The 1st says:

            Although that *is* two games, those are entirely different engines.

            We need the procedural equivalent of Oblivion/Fallout 3 – same base engine, HUGE difference in actual gameplay (Well, in the sense of combat, but still…) – Legend of Zelda: Orcarina of Time/Legend of Zelda: Majora’s Mask – same base graphics, same base sounds, same base gameplay, extremely differentiating in premise, narrative, high-level-gameplay (Young/Adult Link vs. 3-day reset button) -, or…

            Well, just *look* at what the Unreal Engine does. I mean, when you look at that, it’s pretty clear why the Unreal Engine focuses on graphics more than anything else – it sells the engine, which people add gameplay on top of.

            EDIT: Of important note, Lost Odyssey (A JRPG by the guys who used to make Final Fantasy), Tron (Action-Adventure according to Wikipedia), Mortal Kombat (…A side scrolling Fighting game…), Mirror’s Edge (Parkour FPS), AND Bioshock Infinite (FPS). That’s pretty impressive for an engine known for the titular Tournament FPS games.

    • decius says:

      Imagine having an AI card/dedicated AI processor, like a graphics card is today. Instead of having an appearance differential based on player hardware, there would be a gameplay differential.

      Graphical fidelity is a great place for pushing the envelope, because it can be downgraded nearly seamlessly.

      That said, I’d be willing to invest a few hundred dollars in an ‘AI card’ next time I do a motherboard upgrade. (Assuming I can find a mobo with enough pci ports and a beefy enough power supply)

      Hmm… there might be something to that… is there a way to use video card processors to run arbitrary code? Dump the AI code to the video card, send it updates on the situation, and have the vid card make the AI decisions at some level. It would be limited to programs with low-quality graphics, and would only be useful in real-time or time-sensitive applications (the whole point would be to make AI decisions faster, which is hardly needed in a turn-based environment).

      So, I’m basically thinking about using a dual or quad crossfire/SLI setup to update the AI in X-Wing Vs TIE Fighter… and I’m ok with that.

      • Mistwraithe says:

        But 3D graphics is largely a repeatable easily explainable set of tasks and a limited number of sensible approaches to rendering it.

        AI is a much fuzzier concept with adhoc heuristics all over the place. Maybe at some point AI will crystalise to the point where an AI library (DirectAI?) could be put together and accelerated by hardware but I have my doubts.

      • Spectralist says:

        You can use a graphics card to execute arbitrary code. With NVidia it’s called Cuda. I don’t know what it’s called for ATI or if there’s an api that lets you write for both of them at once. You can also use shaders for this if you want. Not really their intended purpose but it is theoretically possible.

        But there’s no real need to. AI is almost never a bottleneck. Outside of path finding throwing more power at it isn’t going to help much very often.

        • Rodyle says:

          Actually, for certain games, such a card would do wonders. You could have an RTS in which every single one of your units has its own AI, and would do shit even if you aren’t around to command them. What about a game like GTA, where an AI card could generate a ton of NPC behaviour so that the streets actually look like there are people living there.

          On the other hand, this could be done just as easily by sticking a few extra gigs of ram into your computer and upgrading your CPU…

          • Alexander The 1st says:

            Legend of Zelda: Majora’s Mask is the only attempt I know at actually trying to make AI that looks like it lives in the world, and they had to use scheduling, and even then – it only works for 3 days at a time before they have to reset it.

            EDIT: Er, let me add to my thoughts. <_<

            In an RTS, you don't want your units doing things you don't tell them – it could easily bug out and send all your worker drones to rush the enemy base if you run out of resources, for example.

            Also, using AI cards would start getting difficult if two game players on separate teams have different AI cards. While it could be argued that a better graphics card would allow for less latency and/or a more effective sniper capability, it still requires the player to get used to it and able to fight properly.

            • Adam says:

              As someone who plays lots of RTS I would LOVE for my units to have some AI. The losses do to the occasional “Run out all the drones” bug would be far less than the current losses due to the “durr I’m a brainless idiot, I’m going to sit here and die” mentality.

          • decius says:

            “this could be done just as easily by sticking a few extra gigs of ram into your computer and upgrading your CPU”

            How is that different from the functionality provided by video cards?

            Oh, right: Shared FSB, L1 cache issues, RAM seek times… Hardware limitations. Good pathfinding would be a start, but my current thought would be “Choose a location to move to and find a route to it based on what you know about enemy locations from direct sight and sound, as well as communications with allies. Avoid moving through areas under fire from known enemy locations. To a lesser degree, avoid areas which could be under fire from an unknown enemy. Running to and then standing on the piled corpses of several headshot allies is right out.

            I’m not sure of the scaling, but I suspect that a large number of weaker processors, each addressing a relatively small amount of dedicated RAM as well as the shared RAM, and each controlling a number of discrete AI elements, would work better than one big processor. Give each ‘unit’ what it needs, let it bang out the decision and tell you what the AI element does.

            The HARD part is when the AI needs to predict the player’s reaction to their actions: Assuming the AI is antagonistic and the situation interesting, the optimum response for the player will be a response NOT expected by the AI. [Set ambush for location A]{send attackers to hide at location B, overlooking location a] [bait enemy towards ambush]{go to position exposed from location A}

            Player: “I see the pattern!” (Moves to location C, ambushes the ambushers, wins.)

            Next iteration: [Set ambush for location C]…{go to position exposed from location A}

            That’s just what I see as the AI goal for cover-based shooters.

            • Alexander The 1st says:

              There’s also another thing to consider with AI, from a diagram I’ve forgotten where I saw it:

              Bad AI ——————————————–Good AI
              Easy AI———————Hard AI—————-Easy AI

              It’s usually not scaling that’s the issue, but that you can’t have AI that’s too good.

      • Adam P says:

        From what I understand, every game has different needs for it’s AI. The only common elements behind AI are few. If am AI accelerater card did exist, then it would either have to do those few things really well on a large scale, or be able to do everything that every developer could possibly want for AI. In the latter scenario, there would be some developers who would only use some of the card functions, making the rest of the functions worthless. And then what happens if a developer wants to do something that the AI card wasn’t made for? Back to the CPU with the calculations.

    • Simon Buchan says:

      Development budget of Modern Warfare 2: $40-50 million. After marketing? (and distribution, etc…, but mostly marketing) $200 million. Most games spend about twice as much on marketing as development, so even if the game would be more improved by another $10 million in, say, AI (somehow), that won’t make the marketing more effective, and that’s what sells the things. If it makes you feel better, game publishers care *far* more about review scores than any other medium I can think of…

  3. Woodthorn says:

    We need more people researching “crouching behind brown chest-high wall” technology! Think of the many areas not explored; Crouching down on your left knee instead of the right, alternating between the two knees, crouching down on BOTH knees. The possibilities are endless!

    • Daemian Lucifer says:

      Dont forget ducking.And various degrees of duck-walking from cover to cover.Leaning and peeking also.

      Unconnected though,did anyone try out the game hydrophobia?It has interesting water mechanics.

      • Simon Buchan says:

        You joke, but cover mechanics have seen a lot of development, and the badly-implemented ones stick out from the well-implemented ones a lot. I’d say that there probably *are* a lot more things to learn about cover mechanics – hopefully figuring out a way around rooms filled with nonsensical chest-high walls is coming (Dark Void had something close, where you could use any wall/floor as “vertical cover”).

        I thought Hydrophobia was OK, I played the Steam release: Prophecy. Then again, I thought Duke Nukem was OK, so I might not be the best judge.
        A bit obvious, but the water was awesome. The protaganist feels well-grounded, with the appropriate amount of “oh shit!”ness (maybe a little *too* much if I’m quibbling), and the antagonist was enjoyably/creepily bat-shit insane. Level design was pretty interesting and they gave you a lot of different gameplay elements to cycle through. Story was a bit silly, but overall served it’s purpose well. It encouraged environmental kills over shooting, with a fair variety to find. I’m also a sucker for The Virus giving you glowey veins – I *really* liked (far more than the Internet tells me I should) Metroid Prime 3: Corruption for it’s subtitular conceit.
        I actually prefered the earlier voice actor for Scoot. What’s wrong with indeterminate British Isles accents people? Disapointed how little time you had with water control, esp. since it’s on the loading screen and everything :/, but then it wasn’t interesting enough as is to hold up over the whole game. The shooting was kinda blah, and they give too much animation priority. Moving in and out of water was overly cumbersome – in particular platforms with water washing over them were a pain. The “we’re all starving!” backstory was criminally badly staged – there is literally negative evidence to that that we can see.

        Overall: Refreshingly different, though not without it’s problems. I’m interested to see the promised part 2, if they ever get around to it, but I can see getting fed up with their story or character control if they don’t step up.

  4. FatPope says:

    Congrats Shamus! You’re in the big leagues now

  5. JPH says:

    You got an Escapist news article about you too!

    Oh wait, somebody already mentioned that in the comments for the last post… Whatever, I’ll link to it anyway.


    I forget how HTML works…

  6. lupus_amens says:


    you are a freaking frontier(hehehe) scientist now.

    • Daemian Lucifer says:

      If only he was doing something about ai instead of procedural world,he could be considered a frontier psychiatrist.

      • Klay F. says:

        The thing is, if Shamus et al. can convince triple-A developers that procedural content is worth pursuing, the average game development pipeline would be drastically changed forever. Instead of spending millions upon millions of dollars on brown spectacle after meaningless brown spectacle, instead those millions of dollars can be spent on making AI not braindead. Just think, a future where millions are spent on bugfixing and playtesting.

  7. Simon says:

    Wow do they have the worst URL ever? .biz sounds so shady as well.
    Is it because of Tim “edge” Langdell?

  8. Factoid says:

    It surprises me how little your hear of procedural content by the big developers. You’d think they’d be all over this stuff as a way to shorten development pipelines, reduce costs, etc…

    It must be some variety of chicken/egg problem. There’s no experts on procedural content because nobody develops procedural content, thus there are no procedural-focused game developers to hire, and the cycle continues.

    Some indie developer needs to make a no-budget mega-hit like Minecraft, but more focused around what the industry is currently doing. Some kind of shooter/action game maybe.

    Then someone will take notice, and maybe that indie developer can convince a publisher to fund them for a medium-budget game.

    Imagine what someone could do with Shamus’s Frontier technology if they had, say, 10 programmers and 5 fulltime artists.

    • aldowyn says:

      The problem is you can’t make an indie game in the style of a AAA game that can compare at all. Any advantages, like the procedural generation, would be drowned out by everything they inevitably did worse. That’s why indie games are always so plain different!

    • Daemian Lucifer says:

      Sadly,it wouldnt work.Spore was a procedurally generated aaa game,and see how no one continued to develop the technology afterwards.And it was made by ea no less.

      • Veloxyll says:

        Spore was also made entirely of mini-games with no binding central game to tie it all together.

        • Daemian Lucifer says:

          While true,it still doesnt matter.It was a procedural aaa game.And not just procedural in its environment,but its characters as well.

          • Klay F. says:

            I think what he’s trying to say is, Spore did not fail because of its procedural content.

            The point is games are not either a success or a failure based on one aspect of its creation. Publishers are the people who like to point at just one thing and dump blame on it. In this case there are almost innumerable thing wrong with Spore, but the procedural content is not one of those things.

        • Alexander The 1st says:

          Going with Daemian on this – it’s not like tech demo minigames with no binding central game has stopped developers from trying to fit motion control gaming into everything and anything.

          • Klay F. says:

            Except by your reasoning, its impossible to make something successful unless you are already a success. Motion controls were only a success because Nintendo basically forced consumers to buy that collection of minigames if they wanted a Wii.

            The reason developers jumped onto the bandwagon so hard (and so late) was because they saw it as easy money. It really has nothing to do with motion controls as a valid form of videogame input.

            • Alexander The 1st says:

              Well yes, they forced it on us. But part of what made that work was that everyone had a standard of controls to work with – Wii Sports, as a tech demo, showed that you could use motion controls in *multiple* gameplay scenarios, and that you didn’t need to re-tool the interface EVERY time you want to implement it. From what I understand of Spore, the procedural generation was different at every level – from bio parts at the beginning, to terrain generation and animation in tribal, to AI and space system generation in the final part – each system was re-written for that specific part.

              The “use motion controls for control input mapping” could let you take the game input you have for one controller, then re-map it without having to re-tool the entire interface to your game. In the same way it’s technically possible to map the DDR-pad to MGS1. You don’t have to re-write the input of MGS1, or the output of the DDR-pad – just get the DDR-pad to confirm which parts are which controls if it was a PS1/PS2 controller.

              Nobody developed on Spore because each level had to be rewritten every time they wanted to do a larger scale.

    • Kyte says:

      I can make a couple of guesses:

      1) Many AAA often have a rigid story/gameplay structure, so they gain little from procedural generation. (Examples: Assassin’s Creed (needs to model the city faithfully), Arkham Asylum (tightly controlled environment design and story flow), Bioshock (also a highly controlled map/env design tied to story progression), Oblivion/Fallout/etc (matching scripted storylines (mainly voice acting) to random generation is kind of iffy))

      2) Most importantly, procedural terrain generation (like all 3D generation) becomes harder the more detail you add. Shamus’s work looks good because you’ve accepted the style and low-polyness. And that’s in fact a good thing. More detail = More chances for the shortcuts to peek through.

      • SlowShootinPete says:

        You might be able to work around the first one by having large regions of procedurally generated areas and drop in some hand-made landmarks for plot-important stuff.

        • Adam F says:

          Also, procedural generation does not necessarily mean “unique for every play through” You can use it once for the original world-gen, and then fix it for every game that everyone plays. It’s not quite as cool that way though.

          • Mephane says:

            Yeah, that’s what FUEL does. The game world is generated procedurally, but all seeds and parameters are fixed, and the developers then added certain hand-crafted the interesting things (camps, racing tracks) on top of that. That said, I still don’t get why people think of FUEL’s world as boring. I think the sheer scale is a real selling point for me, to get a feeling of the different areas to be realistically sized. If it takes like 5 minutes to walk across a desert that is supposed to be like the Sahara, that just doesn’t feel right to me. One solution, of course, are vehicles which bring you around more quickly while maintaining the size of the world.

            And that’s what I like about FUEL the most. I’ve spent far more time driving around than racing, and even when I want to do a race I usually drive to the place myself, and only use the helicopter shortcut to reach the zone’s camp if it is too far away, and then drive from there.

            But I think this is just me. I have a craving for vast open areas in games, for landscape and places to feel like they are actually real-life sized. That said, I can only recommend SpaceEngine. It’s not a game but more a procedural universe exploration thing, but it is the first time I have seen an attempt to convey the vastness of space to actually succeed.

            Long story short: for me not just openness/open endedness, freedom and sandbox gameplay are so interesting, but one major selling point, to me, is scale, both in terms of “the world/universe is indeed a big place” as in “the scales of differently sized things feel realistic in relation to each other”. FUEL does the first thing exceptionally well, while I’ve seen examples of the latter in X3:TC, with really enormous size differences between the ship types and space stations.

        • Viktor says:

          The Elder Scrolls is the perfect target for procedural generation, IMO. There’s no reason they should individually place each hill and stream. Have the program make the terrain, then place cities, roads, and dungeons manually. It’ll be faster in the long run(especially since you can reuse the system for each game).

          • benjamin says:

            I think they are tied for the basic terrain to the maps they made before. but they already use procedural generation for tree.

          • Jeff #3 says:

            The first two Elder Scrolls games used procedural terrain (billed as ‘fractally generated’ in manuals and marketing material) and the results were not pretty.

            Especially the ‘random’ dungeons in Daggerfall.

            Arena had at least a useable map since it was essentially 2D, but in Daggerfall if you didn’t place a mark at the entry of a dungeon, there was a very good change you’d never find your way out. They would spiral around forever, and the exit was some random door in the middle of a gordian knot of corridors.

            • decius says:

              That’s an example of poorly implemented procedural generation. Diablo had procedurally random dungeons that worked out pretty well. The issue is with finding a model for designing a believable ‘dungeon’ that is also playable.

              Or you could procedurally generate a huge world, and then manually add/enhance the interesting parts. I can use Dwarf Fortress (with minimal mods) to procedurally generate an entire world history, geography, and political situation for a D&D game.

              • Alexander The 1st says:

                I can use Dwarf Fortress (with minimal mods) to procedurally generate an entire world history, geography, and political situation for a D&D game.

                Which none of the D&D gamers will read. :p

                Though that could be interesting if there was a seed for it, and you gave it to the D&D players, then they could go around in Dwarf Fortress for extra lore when they want to, rather than listen the GM cite it all the time.

            • Gravebound says:

              That is the first thing you teach a first-time Daggerfall player.

              “Don’t argue, just buy the Recall spell and use it EVERY time you enter a dungeon.” :D

  9. aldowyn says:

    Awesome, Shamus! A lot of what you said is the same kind of thing you’ve told us, but there were a couple new things. It’s just kind of nice seeing a little bit of recognition for the amazing work you’ve done in the past month or two :)

    P.S. Maybe a dev will see that and hire you :P

  10. Kian says:

    “Crouching behind brown chest-high wall” technology is still in it’s infant stages. Notice how many shooters have yet to perfect the mere act of covering the players’ head when they crouch behind a slightly-shorter-than-the-crouching-animation wall! Or solve the Stickiness Paradox, where the player only sticks to cover when you want to run and runs when you want to stick to cover!

  11. Shamus I’ve been loving the project frontier stuff and it’s been making me think “Hmm maybe I should try that 3D stuff again”. Interesting stuff.

  12. Paul Spooner says:

    It was great that they interviewed you and all, but did I detect a slightly patronizing tone? “Aww, look at the indie developers playing with their toys! Aren’t they cute?” The tone of the article didn’t seem all that impressed. I kind of felt insulted, because I thought the procedural stuff you and Miguel are doing is top notch!
    Anyhow, gratz on the publicity! And yes, let’s see more procedural games and less brown cover shooters (and sports clones, and bejeweled clones, and basically everything from EA).

  13. Fat Tony says:

    Shamus it’s be nice if next time you have project frontier open you could fine some nice landscapes and take some large screenshots (1280×1024 pixels)for us to use as desktops etcetra.

  14. Jim says:

    I can you why aaa developers won’t adopt procedural generation in their games: it’s because they have so many artists and designers who are terrified of procedural generation. They don’t want to loose control and potentially loose their jobs. It’s putting too much of the power back into the programmers hands.

    • Simon Buchan says:

      Artists and designers don’t (literally) pay the bills. If publishers or studio heads could make as compelling a product for a 1/10th the cost, you think they woudn’t fire 9/10ths of their staff? Not that I think that’s actually likely, I’d expect similar to current budgets with the better results from successful procedural content techniques to give the best profit.

    • Naota says:

      I’m not so sure that you could ever completely replace artists using technology like this. Somebody still has to make all of the assets that the computer is placing and generating, and if you’re planning to generate the world once and build on top of it then level designers may still be necessary to work out things like the interiors of pre-placed structures.

      If anything this would actually make large, open-world games far cheaper and easier to produce, so smaller teams would be able create them without going bankrupt or spending decades in development hand-crafting a world from scratch. That’s more time and money for everyone to work on improving other facets of the game, artists included.

  15. Topaz Wolf says:

    Still say that I would give money to be able to play with your tech demo. More if you call it a pre alpha and make updates.

  16. I’d like to just point out that what Shamus is doing is making a Procedural Game Engine, while the other guy in that article is making a Procedural World Generator (and he’s got 4 years head more tinkering done obviously).

    A Procedural World generator is not that unique, I forget the name of it but there is a famous terrain software out there that does this.
    But anyway, the point is that a Procedural World Generator allows very rapid world creation, and if you look at that guys site you’ll see that his cathedral can look totally different by just changing some variables.
    Oh and he’s also using voxels (Blade Runner The Game from way back used voxels, I can’t think of any other games right now)

    A Procedural Game Engine however is a tad bit more unique, Spore is the only one that comes to mind. Although some of the 4X space sims out there might have partial procedural game engines (sorry can’t think of any titles in particular)

    The cool thing with what Shamus is doing is that players could literally “share” entire worlds/planets or even an entire universe of systems and planets in just a few KB of savedata, as everything has a seed that is the basis (or a seed based on a seed based on a seed etc.) so it’s very exciting indeed.

    The other guy’s project has similar potential, he could distribute his “world” by releasing the Exe plus a very small gamefile, which is why it makes sense when he mentions the option of running the game in a browser.

    It’s certainly impressive to see this evolving and I can’t wait to see what Shamus does next with his engine, being a developer myself, it’s very nice to “hang along” on the journey like this.

    • Adam F says:

      A Procedural World generator is not that unique, I forget the name of it but there is a famous terrain software out there that does this.

      Do you mean Terragen? I rather like that program.

      Oh and he's also using voxels (Blade Runner The Game from way back used voxels, I can't think of any other games right now)”

      Minecraft is using voxels! Also some similar and related games–infiniminer and Ace of Spades.

      The other guy's project has similar potential, he could distribute his “world” by releasing the Exe plus a very small gamefile, which is why it makes sense when he mentions the option of running the game in a browser.

      If I understand correctly, he wants to host the world online, somewhere with lots of storage space, and let people interact with it over the internet.

    • Gravebound says:

      “Blade Runner The Game from way back used voxels, I can't think of any other games right now”

      Delta Force and the Comanche flight sim series also used voxels. There was a jet sim that also used them (I remember that being a blurb on the case)whose name I can’t recall.

  17. NeesonLiam says:

    Procedural generation of assets (geometry, textures, etc…) is an excellent way of jump-starting asset creation in the game development pipeline, one that many of the triple-As have been doing for years. Offline, it is possible to generate highly detailed models and textures which, in the hands of artists and level designers, are tweaked to become the vast and lush game environments that are packaged with the game. The problem arises when you expect to generate those highly-detailed environments in real-time, procedurally, as the player plays the game. That amount of detail requires an incredible amount of processing, and the more parallel you can perform it, the better. Piling that kind of sheer processing on top of everything else that is going on requires some pretty high-end hardware. You can use frameworks such as OpenCL to offload a lot of the generation overhead to the GPU which, with its highly parallel hardware, is well-suited for those kinds of tasks. However, every cycle that you spend generating terrain is a cycle not being spent drawing all those fancy graphics, so there must be a balance, and this balance point is currently a rapidly moving target.

    Look at Minecraft, for example. The level generation there is very simple, and yet there are still frequent frame-rate tugs as new chunks are spawned. Now imagine trying to pile on the detail, with doodad-maps and ground clutter, denser voxel structures to allow more finely detailed terrain, higher-detail textures complete with normal, diffuse, and spec maps, HDR lighting, AO, bloom, etc…

    There is a reason that most games that push the procedural game engine boundaries these days are indie projects and small hobby things like what Shamus is doing. Their vastly lower graphical requirements simplify the pipeline and reduce the amount of processing required to generate. Cores can be taken away from drawing pretty shinies, and put to work generating big worlds instead. Still, it is a problem that is only going to get smaller with time. It’s not that the big studios are unaware or ignorant of the potential of proceduralization; quite the opposite, in fact.

    As more parallelization becomes available in the average-range machine, even the brown-wall-cover types will begin to utilize the merits of procedural generation in real-time and not just in offline tools. (Mikko Mononen posted slides of his presentation to the Paris AI convention at http://digestingduck.blogspot.com/2011/07/paris-gameai-conference-2011-slides-and.html ; he talked about automatically generating cover annotation for cover-based games. From there, it’s a small step to procedurally generating the terrain then generating the cover data set as well.)

  18. Andrew says:

    It’d be nice if they bothered to, you know, interview you. Instead of just pulling your commentary from your blog. Are you that hard to get a hold of?

    • Shamus says:

      Oh, they did. We had a good conversation. In truth, the similarities between what I said on the blog and on the phone are due to me sticking to a few core points and hammering away at them.

      • Chargone says:

        heh. i’ve been to hear speakers who i’ve previously heard on the radio or tv. you can tell when they’ve covered the same topic a lot because, even speaking out loud and with no notes, when someone from the audience asks a question on a given subject, they give the Exact Same Spiel, word for word, that you heard them give a year ago half way around the world. not to avoid answering the question, but because they’ve answered it so many times :D

  19. So that’s what happened to NextGen Magazine! Why does nobody ever tell me these things!?

  20. HeadHunter says:

    I’m not entirely sure we’ve “solved” pathing. Even highly-anticipated titles releasing later this year still have major problems with this.

    • Wtrmute says:

      Pathing is solved: the A* algorithm is the one everyone, bar none, has been using for (at least) the past fifteen years. The problems with pathing are due to the fact that a path that is calculated as optimal when the mob starts moving doesn’t necessarily stay optimal after a few steps. Real people are constantly recalculating their paths and projecting the movement of other mobile entities (people, cars, dogs…) constantly, which most game entities aren’t willing to do, since it takes too many resources if every enemy runs A* every frame.

      Of course, there’s also the problem that the utility function used by the A* isn’t necessarily very good (often it doesn’t take things like cover or stealth into account) and that makes the pathing seem bad. That’s a more general AI problem, though.

      • NeesonLiam says:

        Heh. You say pathing is solved, then proceed to give a few reasons that show how pathing isn’t solved. There are still a few best-practices to iron out before you can call it a “solved problem”; things like temporary obstacle handling and so forth.

  21. Cezar says:

    Using your procedural generated world:
    1. Get noticed by Edge Magazine
    2. Find a way to get paid
    3. ????
    4. PROFIT

  22. Zak McKracken says:

    Not directly related to this topic, but I didn’t know a better place to put this:

    Procedural content is quite alive, I’d say, it’s just not yet arrived in AAA games.
    Too sad, though, that they didn’t actually make the game this generator was made for. But it’s available as part of the current Humble Bundle, if you want to have a closer look at it …

Leave a Reply

Comments are moderated and may not be posted immediately. Required fields are marked *


Thanks for joining the discussion. Be nice, don't post angry, and enjoy yourself. This is supposed to be fun.

You can enclose spoilers in <strike> tags like so:
<strike>Darth Vader is Luke's father!</strike>

You can make things italics like this:
Can you imagine having Darth Vader as your <i>father</i>?

You can make things bold like this:
I'm <b>very</b> glad Darth Vader isn't my father.

You can make links like this:
I'm reading about <a href="http://en.wikipedia.org/wiki/Darth_Vader">Darth Vader</a> on Wikipedia!

You can quote someone like this:
Darth Vader said <blockquote>Luke, I am your father.</blockquote>