|
|
It used to be a big problem for me to create the huge amounts of volumetric data required every time I wanted to show something new. I have often spent more time setting up generators and waiting for the results than I spent adding new features. The new evolutionary genetic-programming based content-creation pipeline has a good chance to change that once and forever.
While such pipeline doesn't speed up the particular generators nor it is capable to create everything we can imagine, its potential by far exceeds what is common in most computer games of today. It creates unexpected shapes, strange but attractive combinations of elements and beautiful subtle details.
Click the image to enlarge.
The development of such evolutionary generation pipeline can be really painful, so why all the effort? These could be the most obvious reasons:
- tons of fully volumetric composite atom-based physics-enabled objects created automatically for almost free
- content creation easy to combine with compile-time physics-based erosion
- can generate whole levels cheaply (expensive staff focusing only on the important pieces of geometry)
- easy to update/rework whole scenes late in the development
- anybody in a company can contribute to the evolution of each of the levels during the whole development period (staff can rate particular objects and game levels every day); after months or even years of this kind of optimization the results have a good chance to be simply perfect
Evolutionary design can be applied to create many different kinds of geometry – stones, buildings, sand dunes, plants and also whole game levels. And such technology can be running 24 hours a day on hundreds of computers simultaneously. It took me months to develop this pipeline (and it is not over, yet:) and finally it seems that it was worth the effort.
Click the image to enlarge.
You can see various surface details on the images. The material composition formulas (and so the resulting 'depth') have been created using a genetic programming algorithm and applied to a sphere. All formulas have originated from the same initial formula set up by me. These shapes are the result of multiple mutations and crossings done within the first 25 generations. The colors aren't the result of any evolutionary process, therefore they are so boring. Also no fine surface geometry has been applied, yet. The holes in some of the surfaces are irrelevant, the renderer just makes me know that there is no color information associated with some voxels.
Click the image to enlarge.
As you can see on some of the images the actual hi-res geometry (left) isn't always suitable for being used in a product, a resized reduced model has to be used instead (right). That is the basis of super-sampling - a very expensive method to obtain high quality results.
Click the image to enlarge.
These are just the first humble W.I.P. results evolved during the last few days. After the evolutionary design code gets combined with the new content management the generators should be capable to provide me with more data of higher quality than it was the case in the past.
My current goal is to release the first hi-res first-person view example as soon as it is possible. The generators will have to create over sixty times more data than ever in the past, topping at about 4000 times more in case of super-sampled high quality results. Clearly it makes no sense to try that on a single retired computer (it would take months to years) therefore I will have to update few more algorithms before starting the process.
Please be patient.
| |
|
|
