Print

Please login or register.
1 Hour 1 Day 1 Week 1 Month Forever
Login with username, password and session length

[Tottel]

Is there any chance that you'll be at the GDC in Germany next week?

[cameni]

No, not this year yet. Hopefully the next one

[richiebogie]

Hi Brano,

The podcast and this game news is interesting. I have a few questions:

1) Will people be able to create their own COLLADA 3d objects and import them to their piece of the world?

If so, do you then embed that in the server copy of Outerra?

As these are probably not fractal-compressible, I am wondering how quickly you could find the world growing from the base 12GB...?

2) Will you have some village tiles where small parcels of land are sold off for community based housing, shops, schools etc, or can people subdivide their own tiles?

3) I see you have done sand, grass, water, rocks, pine trees and snow. Have you done dry regions like in the Middle East or Central Australia? Eg. Red iron-rich sands of northern Western Australia.

That land should probably be paired with the occasional stunted bush, or maybe a mini-pine tree!  

***

Once the demo / game are released, you will be busy with support, customer feedback, etc. Should be an exciting new direction.

Richard

[cameni]

1) Will people be able to create their own COLLADA 3d objects and import them to their piece of the world?

If so, do you then embed that in the server copy of Outerra?

As these are probably not fractal-compressible, I am wondering how quickly you could find the world growing from the base 12GB...?

We want to enable the import in the game, so that people can test it and test how their stuff will look like there. Whether it will also automatically replicate to others so that they can see the object is questionable, because of the possible legal issues when someone uploads a model for which they don't have the rights to distribute. Usually when you buy a model you can't distribute it in its original format, but when it's converted to a proprietary format from which it cannot be retrieved back, it's usually ok. But it's a bit questionable. In any case, we'll need to implement a mechanism that allows pulling off the infringing model upon request from author. Alternatively there's possibility for the server to function just as a tracker, in which case the actual model would be hosted elsewhere and our server will contain just a link to it, but that's not ideal either.

Will you have some village tiles where small parcels of land are sold off for community based housing, shops, schools etc, or can people subdivide their own tiles?

For the demo game we plan to assign a bunch of tiles to players where they can build their stuff however they want. No sub-parceling in common tiles is planned for the beginning, it's more like that a whole city can be owned by one player. But it may change later.

I see you have done sand, grass, water, rocks, pine trees and snow. Have you done dry regions like in the Middle East or Central Australia? Eg. Red iron-rich sands of northern Western Australia.

That land should probably be paired with the occasional stunted bush, or maybe a mini-pine tree!  

Much more climates and ecotypes will be coming once full climate support is finished. People who then appear in a desert will be probably given a chance to relocate

[richiebogie]

Thanks Brano.

A few questions on the engine.

Currently, is it just world elevation data you pack into 12 GigaBytes of tiles?

Will you soon be packing climate/land class, rock colour, etc?

Are those procedural / fractal calculations mainly used to fill in detail between sampling points? Are they performed during packing or during displaying or during both?

Does the use of fractals mean that viewing-from-a-distance calculations are much more efficient? ie Did that enable the continuous zoom from orbit to ground level? Would you need about 6GB of data loaded to do this? ie not part of the demo?

Would procedural & fractal-generated trees (not just location) be more efficient and smoother to zoom in and out from than using 7 different tree 3d objects with different levels of detail?

Can you pre-calculate or compute on the fly the width and placement of rivers from land contours and catchment sizes?

Sorry if I have exceeded my question limit, but it is fascinating, and I have always loved looking at maps.

[ZeosPantera]

I will try to answer a few of those.

Yes..

Yes..

Maybe..

Yes.. See every video with space to ground transitions  http://www.youtube.com/user/cameni47#g/u

Trees are Procedurally Generated now it has just started testing.. http://www.outerra.com/forum/viewtopic.php?id=390

There are certain resources available for river placement and size and also to smooth the shorline.

[cameni]

Currently, is it just world elevation data you pack into 12 GigaBytes of tiles?
Will you soon be packing climate/land class, rock colour, etc?

Yes, just the elevation. But land class/climate will take much less. The demo will come without them yet, will be added after that.

Are those procedural / fractal calculations mainly used to fill in detail between sampling points? Are they performed during packing or during displaying or during both?

Yes, they compute the missing higher-frequency detail using fractal probability models. They are performed during rendering, the packed dataset only contains the remapped original data (from 90m to roughly 76m)

Does the use of fractals mean that viewing-from-a-distance calculations are much more efficient? ie Did that enable the continuous zoom from orbit to ground level? Would you need about 6GB of data loaded to do this? ie not part of the demo?

The use of fractals has brought in the missing detail below 76m. Continuous zoom is possible thanks to effective LOD management. You don't need 6GB of data when looking from orbit - it only needs data of required level of detail. You would force it to download half the dataset only if you scanned half of its 510,072,000 km² surface in detail sufficient to fetch 76m-gridded level. There's 1,572,864 of these tiles so that's plenty of scanning to do.

Would procedural & fractal-generated trees (not just location) be more efficient and smoother to zoom in and out from than using 7 different tree 3d objects with different levels of detail?

Procedural trees are generated in discrete steps of level of detail as well, what they bring is an automated process, better transitions between LOD levels and more effective rendering.

Can you pre-calculate or compute on the fly the width and placement of rivers from land contours and catchment sizes?

In theory could be possible, but we are going to use the existing vector data for rivers. Smaller tributaries that aren't covered there could be computed using an analysis, but it wouldn't be on the fly, but an off-line process whose output is another vector database.

[ZeosPantera]

See, my response word for word!

[C. Shawn Smith]

ROFL!

[richiebogie]

Thanks guys. Comparing the 2 responses illustrates the level of detail concept quite well!

So is the planet from orbit a sphere with elevations sampled at say 300m squares plus atmosphere?

Any lower uses flat tiles which overwrite the near bits of the low-resolution sphere...?

... Or are there 7 planets worth of resolutions...?

[cameni]

I must say I don't know what you mean.
Anyway. Here's a shot from above with some debug info:



The tiles directly below have resolution of 2.5km/pixel. As you get closer they are being refined, but it's still five more subdivision steps until we get to a tile that runs out of dataset resolution and starts to generate additional data using fractals. However, as you get closer the visible range shrinks and the outer tiles recede behind horizon, and so the amount of tiles doesn't grow exponentially.

[richiebogie]

Cool, I think I get it!

There is one 12GB dataset, but the engine serves up the needed subset of that data as your viewpoint zooms in and out and moves across tiles. In your image above you only need 1 in 33 of the east/west elevation points and 1 in 33 of the north-south datapoints, making approximately 1 in a 1000 or 12MB of the 12GB dataset, but the engine probably sends more data in anticipation of all possible changes in view: north, south, east, west, or change of zoom.

Each tile is flat, but together they form a polyhedron resembling a sphere.

Is that correct?

[cameni]

Nope, tiles aren't flat, their mesh is curved properly to the radius of earth.

[richiebogie]

So I couldn't tile my bathroom floor with them! They're like pieces of orange peel.

I guess I was thinking of a sphere in blender, where objects are ultimately composed of flat polygons... but that isn't needed on this scale!

Thanks for your patience. I will let you do some developing now! I hope I haven't held up the demo!

[richiebogie]

I can see that the server sends a user's computer the minimal data it requires to move from viewpoint to viewpoint. Eg. View from the moon changing to view from the surface.

But does the server require the entire dataset in memory?

If not, how does it efficiently get the data it needs out of files on the hard drive?

Are there separate sets of files for different resolutions? Eg multiple quad tree mapping could map the globe onto 6 files for least resolution for viewing from space, 24 files for the next resolution, 384 files for the next resolution for views maybe 200,000 ft above the surface etc?

For each of these resolutions you would only need to load a subset depending on viewpoint.