It's all in your head, Part 10

[vanarbulax]

Nobody is going to mention that the third screenshot has a person in it? Is it me or does it look out of place in the immaculate and sterile world that subversion seems to generate. Maybe it will look more natural with crowds of people (I'm assuming if there are going to be people there would be a few of them in the city, unless it's meant to be an empty world).
Anyway an easy to use map editor would be great since I would love to try making my own stuff (would still probably turn out crap do) but don't sacrifice versatility just so skill-less people like me can experiment for a few hours. While lots of community stuff to choose from is good quality beats quantity.

Oh and Happy Birthday!

[martin]

I'm still not sold on the idea of a map editor in Subversion though... procedural generation to the max would be better. Why can't I just say "Build me a fancy-shmancy office with creative bits and nice spots for terrorists to camp with AK47s"?

My guess would be that you can design buildings, and the procedural generator will come to place a building and either use yours or use an autogenerated one - so the mission will be set in a user made building with all the carefully designed tactically interesting room, while the rest of the world will be autogenerated to save time.

[Rkiver]

Belated Happy Birthday Chris.

Also ooh map editor, shiny.

[Xocrates]

Even more Belated Happy Birthday Chris!

And I'll go with the rest of the folk and say oooh shiny!

[andrewdoull]

I'm still not sold on the idea of a map editor in Subversion though... procedural generation to the max would be better. Why can't I just say "Build me a fancy-shmancy office with creative bits and nice spots for terrorists to camp with AK47s"?

It's not easy. Procedural generation of interesting 3d spaces is hard. I've written a six part article on procedural generation that suggests how it might be doable, if you're interested (http://roguelikedeveloper.blogspot.com/ ... dural.html)

My guess would be that you can design buildings, and the procedural generator will come to place a building and either use yours or use an autogenerated one - so the mission will be set in a user made building with all the carefully designed tactically interesting room, while the rest of the world will be autogenerated to save time.

That's the approach the likes of Angband take with dropping pre-generated vaults into a procedurally generated world.



Andrew

[Sardtok]

Belated Happy Birthday.

Just heard about Subversion related to the Gamasutra Webcast, haven't really been following the game scene in a long time, so had to check it out.
By the way, 100m x 100m = 10,000m^2
Each square is 1m^2, and each is one quad, which is why you would end up with 10,000 quads each 1m^2, which would be 10,000m^2.
But yeah, if they were each 1dm(decimeter), you would end up with 10,000dm^2, which would be 100m^2 (1m = 10dm. So 1m^2 = 10dm x 10dm = 100dm^2).
I'm guessing you were tired or something while putting together this blog, as I can't imagine that basic geometry would be a problem considering what I've seen you guys release before.

Well, good luck on deciding on a map editing system and stuff, I guess I'll come back here and check out the blog once in a while if I have time.

[briceman2]

RE: Quadtrees & non-Orthogonal lines

Here's a hybrid idea: use a quadtree as your organizing structure. For non-orthogonal lines and lines that do not fall exactly on the leaf node grid, handle them in two steps:

1) do the standard quadtree approximation, but only tag these leaf nodes as containing irregular data

2) from each marked leaf node, link to a list of geometric data describing the segment of the irregular line passing through that leaf node.

So you would essentially be using the leaf grid to chop up irregular lines into segments, then representing these segments as children of the leaf nodes. Access is all via the quadtree, except now the leaves can hold non-standard geometry data in addition to the standard quadtree data.


EDIT: for completeness sake I'll add two points -- but you'd probably think of them yourself.

3) the ideal granularity of the quadtree nodes (for standard data) may not be the same as the ideal granularity for chopping up non-orthogonal lines, planes, etc. One solution is to hang the irregular geometry data off quadtree nodes above the leaf nodes. Then the subdivisions wouldn't produce so many polys.

4) You might even be able to further reduce polys by checking how much extent of a subdivided geometry lies within the viewport -- then render the extent as a single object rather than the sum of it's subdivisions. These "unions" might be pre-built and hung off the higher quadtree nodes for quicker access.

Basically the idea is that quadtrees can store useful data at all levels, not just the leaves, as is standard.


P.S. Happy Bday! Do you get a real cake? Is the cake real? Are you sure?

[RabidZombie]

Happy Birthday!

There's precedent! Precedent, I say!

Oh, blame it on me, why don't you...

Nobody is going to mention that the third screenshot has a person in it? Is it me or does it look out of place in the immaculate and sterile world that subversion seems to generate. Maybe it will look more natural with crowds of people (I'm assuming if there are going to be people there would be a few of them in the city, unless it's meant to be an empty world).

We're so used to Doom sprites as placeholders I guess all our minds just blocked it out. (It might be a placeholder.)

[MarvintheParanoidAndroid]

I'm in no way any kind of modder or mapper, so a usable map editor like this looks awesome. I've played around briefly with the map editors in Red Faction (waaay back) and UnrealED, and I was reasonably surprised that I could at least manage to link a few rooms together, but further than that I definitely couldn't create anything that would be particularly fun to play in. And Hammer completely terrifies me, so since switching to mainly playing Half-Life 2 and its mods, I haven't really gone any further in any of my short-lived mapping attempts. Anyway, I'm definitely looking forward to having a go at using this!

[Cooper42]

Many Happy Returns

I'm excited by the prospect of user-generated content, which, as far as I can tell, is the only clear focus of subversion so far...

In which case, yes, it needs to be a very easy map editor. I used to love the old Dungeon Keeper map editors, which were followed eventually by an official one. One I remember was text based and way beyond me, the other was a kind of wysiwyg, very simple but I remember being able to put together skirmish maps very quickly and easily. That was good fun. Some of the WAD editors were good too, others just baffling. Now I look at things like Hammer and I wouldn't know where to begin.

If the plan is to keep with the 2d-plans of buildings, you've saved a lot of hassle right there. Even a Doom pseudo-3d type of format would be good, it's the kind of format which people can wrap their heads around pretty quickly. A half-way house for those of us with less time/patience/technical knowledge would be appreciated.

What I always get worried about is the scripting, if that's going to be available for user-generated content. Unless the scripting is in simple, game-related terms (ZZT-OOP anyone?), it bypasses many people...

[xyzyxx]

...user-generated content, which, as far as I can tell, is the only clear focus of subversion so far...

I thought the clear focus was on procedurally generated content? Surely user-generated content will be secondary to this.

[byrneda]

Hi Chris - Belated Happy Birthday btw.

I can't believe you mentioned Dwarf Fortress in your blog, but it makes sense, from a procedurally generated aspect. I'm only flabbergasted, as having messed around with DF's rendering program, I'm amazed at how much is randomly created (eg. a thousand years of history randomly generated). Then there's the initial time period of waiting for the environment to be created (rather slow on a Centrino Duo laptop).

Which asks the question, if a full workable and usable environment takes minutes to render, how will this reflect on the gameplay of Subversion? Will you have a classic "please wait.. loading screen" and be able to then "access all areas" or will it "quietly" render the environment in the background based on the Quad Tree and the LOD set from the user's viewpoint?

And what about using Subversion in a multi-user environment. especially if Subversion allows the user to edit the environment on the fly.
I've played cube, and I was amazed at how easily accessible it is (via the menu and its way of displaying new levels etc) and I didn't know one could edit the environment - I was wondering why I started seeing box-points in the buildings.

I know its way too early to tell how these things will pan out, but I thought I'd mix it up a little and see what everyone has to say on the subject

Dang, I'm a sucker for all the nitty gritty details - probably why I must thank you Chris for letting us see what you're doing and how you do what you do.
Reminds me of the days when you could "finger" John Carmack's .plan and see what's new:
(shame, it don't work)

http://www.shacknews.com/finger/?fid=jo ... ftware.com

P.S. Having seen this video on Hammer http://www.youtube.com/watch?v=hQsXEoQRmSs I have to say that it reminds me of 3d Studio Max and Maya (both of which I've never really used fully. And talking about getting bored with something, I even bought a training video for Maya and ended up thinking - "its just like 3D Studio Max and moved onto my next biggest love - playing games).
I'm all for simple-to-use editors, and love the idea of procedurally rendered landscapes. Have you thought about using something like Google's Sketchup? I suppose you could place a wireframe at the ground level (which could be configurable to use different wall widths) but be just as easy to extrude with.

I'm reckoning the user aspect of procedured landscapes would be like the artist with a blank canvas, or that kid with the spray paint in his hands. The ability to fine-tune or simply paint the environment you want on the walls that have been rendered for you.
(One of the things that still gets to me about Google Sketchup is that if you kill a certain node point, you lose the face of the wall and end up with a nasty object that has no wall.. - I think this idea of Chris' would be a definite step forward, where you magically create walls - you don't need to worry about these landscape tears).

And lastly - scripting. Maybe my idea of scripting (think of batch files) is different from scripting in games - can anyone point me in the right direction? I know of Ruby and Python, but I don't know where scripting comes into the programming aspect as most games I've played aren't that easily accessible to the scripting language (unless you want to go hardcore - nope, not pointing fingers at UnrealEd or anything).

Maybe that procedurally generated building idea of mine using simple words will someday come to fruition

[zoombapup]

One of the things we tried in Worms 3D Chris, is to use skinned voxels. We were playing around quite a lot with different terrain systems (worms had to have destructable terrain didnt it ) and Andy was playing with something like 6-8 layer intersecting heightfields. I made a suggestion that what we needed was volumes, but ones which were on a non regular grid.

We tossed around some ideas about localised volumes and Andy went away and prototyped it. Took a day or two (Andy is a wicked fast prototyper).

Anyway, what it did, was to skin the volumes in realtime. So you could destroy the volumes (think of the world as a volume of sub volumes, each of which has its own local space) and it would then skin the volumes using something akin to metaball code.

It got marketing branded as "poxels" (for polies + voxels), but weirdly enough, once the art and level builders got into the idea of making solid volumes and then manipulating them, it was pretty damn powerful and the WHOLE world could be destroyed.

I think your Quadtree approach works well though, because for the most part cities in modern countries are layed out on grids and that will help with the render speeds.

One thing you might think of while implementing buildings btw. I was going to submit a paper on this in GPU Gems. Think about the possibility of rendering buildings from floorplans and how similar the plans are? Why not use instancing??
My plan was to create essentially a single "full" floor, that is, a floorplan with every single element in it stuffed into a set of VB's. Then I was going to pass per-instance data into the geometry shader to allow it to "drop out" some of the instance geometry. Essentially taking a full featured level of a building and taking parts of it away (walls, doorways etc). That way, you simply build one giant set of data and let the geometry shader take care of the details. Assuming you can lay all your data into vid mem, should be screaming fast.

Of course you'd have to mix up a few instances. But given a lot of buildings are very much the same, you could definitely get a pretty fair set of coverage.

Alas, no time right now, got more important AI things to be getting on with.

Nice blog.

Phil.

[elexis]

I think that there should be some sort of upload/download function built into the game that allows you to very easily share your creations with others. Not a website where to upload something you have to go into the game directory and find where the file is, but a one-click friendly in-game interface. Similar to what you will be able to do with your civilization in Spore when it comes out.