Minecraft and Castle Story. Voxel maps?

[flash=640,360]http://www.youtube.com/v/qqDSh34VoPY?version=3&hl=en_US&rel=0[/flash]

At 30 minutes neutrino talks about procedual map creation and terrain generation. There will be no heightmaps but instead map generation will be taken to the 'next level'. With games like Minecraft and Castle Story mentioned, this sounds a lot like you want to do something with voxels.


What the heck is a voxel and how does it look like you might ask. I guess most of you have a fair idea how Minecraft looks and plays (shame on you if not), so here is a short demonstration how Castle Story's maps look like:

[flash=640,360]http://www.youtube.com/v/NwZdXRPeYrc?version=3&hl=en_US&rel=0[/flash]

Now my question is: Is my assumption correct and are you guys testing if voxels can work for the map generation?


I have worked a fair bit on maps for SupCom and FA and have a good understanding of how heightmaps work and the limitations they have. I'm thrilled to think what options would open up if you guys could use voxels instead!

Should I start preparing my tools?

We're still too early to say what we'll be doing. The short answer is no, we are not currently testing voxels. Longer answer is maybe, we're still not sure exactly what direction we'll be going with the terrain.

What we want to achieve is a system that allows for fundamentally different terrain than what's been seen before from randomly generated worlds and RTS games. The gameplay visualization is a low benchmark we plan on surpassing.

Yeah, I assumed the Minecraft comparison was in reference to the greatly varied terrain and use of biospheres to govern unique looks for each climate.

Even if the quality of the demo was final, I would be happy.

After the prototyping, I would really want to know what you went with.

Procedural generation is my personal area of future expertise I hope.
More specifically, the procedural generation of terrain is one of the things I am going to research the coming months.

Check my portfolio in my Sig if you want /shameless,butontopic,plug

The references to minecraft/castle story were really about procedural generation. A voxel approach was considered early on but doesn't really give the look we want. I have some research to do on my current idea. If that doesn't work out then we'll try some other ideas. I really do like the idea that you can carve into the terrain in a way that makes sense. Trying to stay away from a "traditional" height map.

Yeah, the demo really looked quite nice.

Although, it maybe was too... Unrealistic. Sort of. But I can't really explain what I mean. The world just didn't look like a proper world, it looked too... Hmm, let me get a picture.

The planet (to the left, obviously ), has many... Eh, what's the word? You know, fluid 'tentacle-like' stuff on it. Which looks a bit weird. I hope you get what I mean.

It doesn't matter much - I just don't want all planets to be like that; some should be massive pangaea's, others should be archipelago's, others should resemble our Earth, others should be fantasy-like with 'ubelievable' quirks like this one...

Still, what I've seen so far looked plenty nice!

http://www.youtube.com/watch?v=_CCZIBDt1uM

Voxels have tons of potential. Building an engine on something like that reminds me a lot of the TA:Spring project.

That said, my heart isn't set on any exotic tech or destructible terrain. If something like that is going to bloat costs or man-hours then I'd much rather just have standard RTS terrain and solid TA style gameplay. The terrain I see in the demo looks fine to me - though I welcome pleasant surprises.

Thanks for the answers!

The traditional heightmaps have served me well in SupCom/FA (add erosion!) as mapper but there are always some limitations you can't get around easily. I'm looking forward to seeing what you guys come up with.

The Atomantage engine is an awesome bit of tech, it's also not really useful in today's world. There are always trade offs with various real time technologies and many of them can be distilled down to trying to find the balance between computational power and memory usage. Voxel engines generally fall heavy on the memory usage side. Minecraft for example has extremely simplified graphics yet can use several gigs of memory. On the Xbox the terrain is limited to a sometimes annoyingly small 1024x1024x1024 block area purely because of memory constraints on that system, and on PC it's not unusual for it to be able to bring machine with 8 gigs of RAM to its knees. Something like a small Atomantage level can use 4-5 gigs in a heavily compressed format.

Several other games have used voxels in some form though. Crytek uses the tech to be able to carve in to otherwise normal height map terrain, though the resolution of the voxels is actually lower than even Minecraft and the data is used only to generate a smooth wireframe mesh rather than trying to render them directly.

To expand on this, this also explains why the 'infinite detail' voxel world (here) is "fake". They have to use the same 'scene' (or composite of voxels forming a block) that you can visible see tiled over and over. You would never be able to create a dynamic, living world like that with voxels, not in the immediate future.

Infinite Detail isn't fake, it's just the same kind of talking up you'd get when going to a used car dealership. Would you rather have that 1991 Toyota for $300 that will go for another 300k miles or the 1994 Cadillac for $3000. Ignore the hood is welded closed and the back half of the car is a different color from the front, it's got leather seats and a button labelled AC!

We've had 30+ years of polygon raster rendering and we've solved a lot of the problems with using them for real time gaming. Voxels have had significantly less research in this regard. Usually they've been used for static data (visualizing MRI scans for the medical field) or computation of complex systems (fluid dynamics). One of the great unsolved issues with voxels is how to animate something as simple as a vehicle or a character.

Yeah, that's why I had quotes around "fake". It's not just a CGI 'gotcha' video, but what they're promising is (with currently technology!) something of a pipe dream. Notch broke it down pretty well, that unless you plan on tiling the same features over and over, the memory required to simply store the coordinates of each voxel would cost terabytes of RAM if you wanted to make something approaching the size, scope, and uniqueness of a modern game.

It's nice that they're showing what could one day be possible, but the other side of the coin is, who the hell is going to want to spend their time making rocks detailed down to the centimeter level of detail, to say nothing of the rest of the art assets? Unless I'm missing something there, you'd need an art team the size of a small city to pull that one off.

Terabytes of storage, but not necessarily working memory. Euclidean is claiming to have this awesome search technique and it could conceivable require only (several) Gigabytes of data in memory, while using lots of streaming and out-of-core techniques. Even if you need 8 or 12 GiB, it's at least feasible, but not downright impossible.

Still, not saying Euclidean does not look a little shady, but I would definitely not dismiss it right away.


Also those are Point Clouds, not Voxels ;P. While the distinction is unimportant in some cases, I still want to make it here.

Euclidean's Unlimited Detail is quite explicitly voxels, specifically sparse octree voxels, regardless of how they sidestep the word by tossing around "point cloud" or calling them "atoms". Voxels are just a form of point cloud data, as are mesh vertices and any kind of particle effects system used by games and movies today. Voxels are a point cloud arranged in a regular grid in 3d space, essentially a 3d bitmap. Sparse octree voxels are useful for quickly traversing voxel data, useful for rendering, they are also useful for greatly compressing voxel data so that empty space is nearly free ... which I believe is one of the claims made in the Euclidean videos.

edit: Also streaming content is nothing new, nearly every game uses more storage space than available memory. id's latest game Rage made a big deal about their mega texture system, and while their implementation is impressive and novel the idea has been in use for years. If you think of voxels as 3d textures then you can stream them in and out just like games do with mipmap levels of 2d textures. Nothing exciting or unique about this either.

I got 16gb RAM now in 2 years that will be 32gb then 64gb, 128gb etc. Just have patience xd. Look at how the games looked 10 years ago. Over 10 years there will be a huge difference again. With ray tracing, voxels and more stuff normal pc's currently cannot run.

I'm glad to see that Euclidean looks into voxels and their viability. Sure, it might not be something we will see in the next 5 years but it is something worth exploring. Same as ray tracing. At some point it will be possible to do it in real time on your PC at home. Would be a shame if nobody would have looked at it until it is. Voxels are awesome for terrain generation and hopefully we will see progress in that area soon now that games start using them.

This makes it sound like people aren't looking in to voxels, they are. My problem with Euclidean is they're where voxel tech was over a decade ago and making claims about pushing the technology forward when the only real difference is today's computers are faster and have more memory.

In 1996 I had a tech demo I got online running on my Pentium 90 with hierarchically animated voxel characters rendering at 30fps. Through out the late 90's we had games with destructible voxel landscapes, moving voxel objects in 2.5d games, and even fully animated voxel characters in the old Blade Runner adventure game. None of these are things Euclidean's been able to show in their tech.

And Euclidean's last big technology "reveal" was a polygon to voxel converter... How do you think people made voxel objects back in 1996?

http://www.youtube.com/watch?v=dZLEvFqrKcg