Polyphony Digital Presents Procedural Landscape Generation Techniques at CEDEC Event

This really isn't new technology. About 25 years ago I was doing a lot of 3D architectural modeling for animated walkthroughs and static photo real images. One of the tools we used was a program called Trees Pro. You entered some basic parameters to define the type of tree and and the modeling resolution and hit the go button. The program would then produce random procedurally generated 3D trees for use in our models.

Also for procedurally generated 3D landscapes there was a program called Bryce which we put to very effective use on occasion.

Back in those days, however, it would require an overnight render to produce a single typical GTS frame.
 
This really isn't new technology. About 25 years ago I was doing a lot of 3D architectural modeling for animated walkthroughs and static photo real images. One of the tools we used was a program called Trees Pro. You entered some basic parameters to define the type of tree and and the modeling resolution and hit the go button. The program would then produce random procedurally generated 3D trees for use in our models.

Also for procedurally generated 3D landscapes there was a program called Bryce which we put to very effective use on occasion.

Back in those days, however, it would require an overnight render to produce a single typical GTS frame.
Indeed, but PD isn't talking about procedural generation of a tree, or even some trees. Nakatomo's presentation is about the procedural creation of vegetation on a landscape from seed.

The team already uses tree creation software (Nakatomo cites L-System and SpeedTree; I don't know which they use) to create and edit trees, which it then manually places onto its landscapes. This method grows trees en masse across terrain from before the seeds even germinate - in fact it may not even germinate some seeds, depending on temperature and moisture.
 
Indeed, but PD isn't talking about procedural generation of a tree, or even some trees. Nakatomo's presentation is about the procedural creation of vegetation on a landscape from seed.

The team already uses tree creation software (Nakatomo cites L-System and SpeedTree; I don't know which they use) to create and edit trees, which it then manually places onto its landscapes. This method grows trees en masse across terrain from before the seeds even germinate - in fact it may not even germinate some seeds, depending on temperature and moisture.

To be honest it is kind of a natural step forward for, and a convergence of, the technologies we were using a quarter of a century ago. A lot of the stuff talked about here was around at the time, but existed as completely separate entities. For example biologists had access to pretty accurate simulations that would predict the growth of vegetation given various environmental parameters, albeit without the pretty pretty we have today.

I'm not saying it isn't exciting, as it certainly is exciting, and while I may appear dismissive of it in saying it's just a fusion of a whole bunch of pre-existing technologies it's actually fascinating to see how they are being brought together, and how, by leveraging todays computing power and the recent advances in AI, such detailed and natural models can be produced.

The only thing I ask is that when it comes to procedural track generation is that they don't show the AI any Hermann Tilke tracks ;)
 
This really isn't new technology. About 25 years ago I was doing a lot of 3D architectural modeling for animated walkthroughs and static photo real images. One of the tools we used was a program called Trees Pro. You entered some basic parameters to define the type of tree and and the modeling resolution and hit the go button. The program would then produce random procedurally generated 3D trees for use in our models.

Also for procedurally generated 3D landscapes there was a program called Bryce which we put to very effective use on occasion.

Back in those days, however, it would require an overnight render to produce a single typical GTS frame.

I haven't thought about Bryce in a super long time. Forgot it existed, actually. That was a really cool piece of software.
 
Always interested to hear about these type of things, gives me great insight on what devs are working with to continue to enhance my experience as an end user.
 
I want that ! not a tokyo like, I want a London GT6 !
 

Attachments

  • pd-proceduralgeneration-city001-1200x675.png
    pd-proceduralgeneration-city001-1200x675.png
    119 KB · Views: 61
So no course maker / track editor for PS4 YET. And it seems more and more likely we won't see it till PS5.

Ehhh I can't wait.
 
So no course maker / track editor for PS4 YET. And it seems more and more likely we won't see it till PS5.

Ehhh I can't wait.
PS4 is based on 2013 hardware we wont see anything mindblowing now its all about the PS5,faster loading will be great for simracers
 
PS4 is based on 2013 hardware we wont see anything mindblowing now its all about the PS5,faster loading will be great for simracers
For sure. I've wanted GT5's course maker to return with updates at least. It would be a little out of place, yet I still would've welcomed it. Tokyo Bay alone would be awesome :D
 
People seems not so interested in such academic contents, while they may reflect what pd is working on...
Maybe because they are written in Japanese, but I read and write Chinese so it's much better.
God same it, I like the academic institute smell of pd. The same for the two presentations last year.
Good job Polyphony Digital, good job Sony.
 
Did someone say technical documents?

genie_wakes_up_in_potato_quality.jpg



Forward, by me. On the topic of procedural/parametric generation, and these tools in general...

Generation tools are an invaluable asset in modern game-making. Instead of doing things by hand, we can use scientific reasoning and modern computer power to do a lot of heavy lifting. These tools come into their own when making large amount of assets/content/area. The cost can be huge; you can spend months on a tool, and it can all be worth it.

Think of it like this; you can have one craftsman, with a hammer. He can hammer out body panels, and make one car a month. Cheap, and good if you only want to make a few things. OR you can invest a huge amount and build a car factory. It costs millions, takes hundreds of employees to work, but you can now build 1000 cars a day. These tools are like building a factory.

They are a huge strength for the wielder. The ability to create things with a snap of your fingers is great, and can become a defining aspect of your game, or a very handy backup tool to help carry other aspects. Even simple ones can save huge amounts of time; in this case you're looking at how they spawn trees for background landscape renderings (like at Dragon Trail) and close to the track. Rather than wasting a year and 3 interns, you spend a few months with a programmer to make this, then get the computer to do it in 10 seconds for each new track/background.



For the Nagatomo presentation:

P1: opening blurb, title pages.

P3: (extended with info by me) Tree generators have traditionally been one of the strongest cases for automated computer generation. Because they naturally behave like fractals; growing in a repeating pattern.

You can see this in the L-System image on the left; each branch has a random angle offset, but will split into 2 shorter branches. Then this repeats. Looks like a tree eventually.

Modern systems like SPeedTree are considerably more complex, but are also just expansions on that same formula/idea. If you did fractals at school, you know this.
trees_page.png


P5-8: (again, extended by me). More modern systems on powerful hardware can go beyond that. In concept, trees don't grow in isolation, spaced evenly across the terrain. We want to spawn them like they would grow in real life, and create a realistic pattern. It needs water, light, and sediment.

P9: how do we do this?

P10-16: the mice represent the uneven-ness of nature. Once one has an advantage, it out-grows the other, can consume more, and grows more, and so on. Advantages grow and stretch more and more as dominance is established.

P17 - P20: How do we learn to program this? Research sites.

P21-32: (not direct translation). If you're familiar with "The Game Of Life" simulation, you know this bit. The rules are that where conditions like food and a minimum population are reached, a new life is spawned, and if the conditions are too many lives adjacent, a life is removed. This creates repeating patterns of life that spread outwards with peaks and troughs. If you're interested, just google Game Of Life Simulation, Wikipedia it.

P33- P37: the rules and (some) code for the simulation. Note the use of a visual shader editor (looks like a flow graph). These are common for tools like this since they're both more friendly to non-programmers and often easier to debug and trace information / values.

P38 - P42: Height, Height > colour, Light @ summer, Light @ winter, water all displayed. These would be available in inputs to the system.

P43 - P47: Note the trees spawning! they like to be away from the cold mountaintop, and near areas of water.
trees_near_water.png


P48-P49: The larger trees are strangling the growth of the small ones! Note the lack of medium-sized trees; either they get enough to grow massive or don't, and stay small and/or die.
tree_strangling.png


P50 - P52: Finally, good results. Trees grow in a believable pattern!
tree_position_good.png


P53 - P54: references, thanks, etc...

My Notes on the presentation: This is a good intro into the thought process behind developing such tools. The example is simple, bot logically and programattically, but also offers a near-infinite room for expansion.

You might want to do multiple passes with this system; one for trees, one for shrubs, etc, all with different values for how sensitive they are to light, temp, slope, shade from larger trees, etc. That's how you'd get a mountain covered in vegetation looking realistic.

EDIT: I'll do the other one a bit later. It's (saito) not as good as the tree (nagatomo) one; less pics of intermadiary stages. It's on building creation so I don't know why they wouldn't include those.
 
Last edited:
PS4 is based on 2013 hardware we wont see anything mindblowing now its all about the PS5,faster loading will be great for simracers
This tech is more about development and uses server farms to calculate it. The power of a console have nothing to do with it or an track creator.
 
To be fair to PD, they have a history of incorporating 20-30 year old "research grade tech" and putting it to good use. That is probably about the sweet spot for games given their iterative and interactive nature.

Saying that, check out the dates on the references at the end of Nagatomo's presentation!

PDpres.png


"Old" science does not mean irrelevant, certainly.

On the subject of simulated systems in games, the daddy of them all has to be Dwarf Fortress. The world generation is an utter marvel.
https://www.polygon.com/2014/7/23/5...ush-your-cpu-because-creating-history-is-hard


I'm personally glad PD is on the systems side of the great content divide.
 
Stop this graphics things.
Fix the damn SR rating and penalty system.
I’ve been S-rated forever but one day I fell to A and now I’m falling. The game is only playable if you stay in S-rating. If you start to fall there is no bottom and no way back up. People around you start missbehaving really badly and theres a 50/50 chanse you get the penalty for every encounter.

I loved this game but it’s completely broken for me now. And the upcoming nations cup and manufacturing series are allso ruined because of this.

Reinstalling Project Cars or something. I’m heart broken.
 
Back