I'm really looking forward to No Man's Sky and will probably buy a PS4 just to play it.

I am a fan of procedural generation. I am old enough to have played Elite on an Apple II and been blown away by it at the time. (For younger readers, No Man's Sky is basically an elaborate modern reboot of Elite)

However, I know as a developer and game player in his 40's that there is a reason everyone doesn't use procedural generation for everything and I know that we all were more used to putting up with repetitive gameplay in the 8-bit era. Procedural generation is great for demos but is hard to get right for games. Demos need to look cool. Games need to be fun. It is hard to procedurally generate fun.

Another personal data point for me is how dull I found Sid Meier's Pirates! on my iPad even given that I remember loving Pirates on the Apple IIgs in another life and given the fact that the iPad version is pretty much a straight port of the original. Pirates was not exactly procedurally generated -- the map was fixed on the realworld carribean -- but it was an open world game and which countries were at war with which when and where treasure was located etc was randomly generated. I remember I used to like sailing around and doing pirating and trying to maximize yields of each voyage and plan out where I wanted to go, etc. Recently I played the remake a little but had a feeling of "Is this all that it was?" So maybe my tastes have changed, but hasn't everyone's?

It's a good question. I think it looks amazing and it's a great technical achievement to my mind, but I also wonder about this.

I think part of the thing you get in games is knowing that another human being had to put hours of effort into something. That seems to me to be part of the buzz. If this is true then no matter how clever procedural gets, it will never be able to produce the same feeling.

Here's hoping my theory is wrong.

someone said my "fauna vst" was used to generate source for some creature sounds, i'd probably listen over your shoulder for a few minutes :)

And then there's Elite Dangerous.

Anyway. Procedural generation isn't always inherently boring, as it's the main draw for players of the roguelike genre. The Binding of Isaac and Spelunky are quite popular. I think the uncertainty rises from the fact that procedural generation isn't as common in the space game scene (could be wrong).

And then there's Elite Dangerous.

That's one of the reasons I'm cautiously optimistic about No Man's Sky - Elite is a beautiful game, but when you get right down to it, it's an empty wasteland with minimal content.

No Man's Sky has taken an interesting balance of procedural generation and artist created content - the massive variation achieved by mix-and-matching pieces of real-world Animals, semi-realistic terrain biomes, and carefully segmented spaceships, makes the sections of the universe we've seen so far appear to be teaming with life. That's something that Elite really misses on, though they may improve in this regard when/if they move on to implement planetary landing.

On the flip side, the more actual gameplay footage I see of No Man's Sky, the less impressed I am by the technical achievement. Yes, he's written a pretty fast procedural terrain generator, and his marching cubes or what have you is surprisingly seamless at converting voxels to meshes at playable framerates. But beyond that... If you look at screenshots from high orbit, its apparent that there are no macro rules to terrain generation - it's just high-frequency terrain features (which look great from the ground, but uninteresting from the air). And the creatures/ships are literally just mix-and-match of pre-made pieces, plus a few colour and texture variations.

I don't mean to knock their achievements: many kudos to them for building a shipping game and obtaining funding from a major platform gatekeeper! But from a technical stance, even people on this forum have produced more varied and interesting procedural content.

And then there's Elite Dangerous.

That's one of the reasons I'm cautiously optimistic about No Man's Sky - Elite is a beautiful game, but when you get right down to it, it's an empty wasteland with minimal content.

No Man's Sky has taken an interesting balance of procedural generation and artist created content - the massive variation achieved by mix-and-matching pieces of real-world Animals, semi-realistic terrain biomes, and carefully segmented spaceships, makes the sections of the universe we've seen so far appear to be teaming with life. That's something that Elite really misses on, though they may improve in this regard when/if they move on to implement planetary landing.

On the flip side, the more actual gameplay footage I see of No Man's Sky, the less impressed I am by the technical achievement. Yes, he's written a pretty fast procedural terrain generator, and his marching cubes or what have you is surprisingly seamless at converting voxels to meshes at playable framerates. But beyond that... If you look at screenshots from high orbit, its apparent that there are no macro rules to terrain generation - it's just high-frequency terrain features (which look great from the ground, but uninteresting from the air). And the creatures/ships are literally just mix-and-match of pre-made pieces, plus a few colour and texture variations.

I don't mean to knock their achievements: many kudos to them for building a shipping game and obtaining funding from a major platform gatekeeper! But from a technical stance, even people on this forum have produced more varied and interesting procedural content.

Yeah, from reading the New Yorker article I gathered that the planetary surface generator is mostly a collection of hacks rather than an attempt to simulate the processes involved and "grow" a planet the way, I gather, Dwarf Fortress does. He seemed to imply that everything has to work locally like he only generates at fine detail the parts of the planet that the user sees so anything that naturally wants to be global necessarily has to be a kludge. The New Yorker article, i think, discussed rivers not being done the right way. I get the feeling he has several dozen planet templates -- ice planet, grassland planet, and so forth -- that get picked and plugged into when you fly there.

I just wonder how old it will feel once you have seen all the templates. In the New Yorker article though he seemed to indicate that the fun was in the exploring. Like at one point he takes the contoller off the writer of the article because he wanted to check out some caves commenting "I haven't seen caves like that before" but I don't see how it can be interesting to him honestly. I mean isnt there a fixed set of things that can be in the cave or nothing is in the cave but in either case wont one cave be the same as another?

I'm still intrigued though. I mean I am excited about this game.

I just had a thought, once the game goes Live, changes to the procedural algorithm will be... challenging. Maybe they could maintain several versions of generation algorithms for use with visited locations along with the latest one for those yet to be generated.

Yeah, from reading the New Yorker article I gathered that the planetary surface generator is mostly a collection of hacks rather than an attempt to simulate the processes involved and "grow" a planet the way, I gather, Dwarf Fortress does. He seemed to imply that everything has to work locally like he only generates at fine detail the parts of the planet that the user sees so anything that naturally wants to be global necessarily has to be a kludge. The New Yorker article, i think, discussed rivers not being done the right way.

This has been one of the biggest roadblocks for my own large-scale procedural generation. The easiest types of procgen to stream are those that are represented by localized mathematical functions: Perlin noise and other similar constructs. No large chunks of content to store and transmit; just the elegance of a function. You can pass an entire world encapsulated by a simple seed, and any given chunk of the world can be generated in isolation, without requiring any knowledge of neighboring areas.

Unfortunately, such methods don't square very well with physical realism. Natural features such as rivers are direct results of processes requiring in-depth knowledge of neighboring areas. A river flows from somewhere, to somewhere, and how it flows is a function of how one given physical location corresponds with another. It's not a problem that can be easily solved through a pure mathematical function alone, but typically requires large-scale iteration over entire chunks of terrain. This complicates the process of generation chunks in isolation, because each chunk is dependent upon neighboring chunks, which in turn are depended upon their own neighboring chunks, ad infinitum.

The best you can do is try to find creative mathematical constructs whose knowledge of the neighboring areas is implicit within the structure of the function. Much like Iñigo Quílez's work with derivative noise, where the derivative of a particular function layer is used to inform successive function layers, thus incorporating a derivative's "knowledge" of the surrounding area into the process. I remember Dmytry posting a tantalizing thread regarding modeling erosion fractally, then disappearing for a decade or so without (afaik) providing any kind of how-to. His page has some tasty images,t hough, and maybe now that I'm thinking about it again I should dig around his stuff some more.

I just wonder how old it will feel once you have seen all the templates. In the New Yorker article though he seemed to indicate that the fun was in the exploring. Like at one point he takes the contoller off the writer of the article because he wanted to check out some caves commenting "I haven't seen caves like that before" but I don't see how it can be interesting to him honestly. I mean isnt there a fixed set of things that can be in the cave or nothing is in the cave but in either case wont one cave be the same as another?

I'm still intrigued though. I mean I am excited about this game.

I've worked quite a bit with various procedural generation schemes, and I am still occasionally surprised by "happy accidents" generating some particular formation or layout I hadn't anticipated. Much like I am still sometimes surprised and amazed at some of the formations constructed from the Mandelbrot set, even though I've written numerous Mandelbrot set visualizations before and understand the nature of the process. Still, eventually you do have a deep enough understanding of your own algorithms that such surprises become somewhat less frequent. And for every "wow, haven't seen anything like that before" moment, you have thousands of "yawn" moments as the generator builds pretty much exactly what you expect it to construct.

The E3 "gameplay" demo was boring. Despite the impressiveness of the technical scope of making a galaxy that appears to be teeming with life, there is a distinct lack of concrete objectives. The game may end up being a sandbox with dozens of available, randomly-generated encounters. Maybe that's enough?

The game may end up being a sandbox with dozens of available, randomly-generated encounters

I put 50-60 hours into StarBound back when it was in alpha, and that didn't have a whole lot more to offer, objective-wise.

I think a sandbox 'tricorder simulator' is enough for a lot of people, assuming the content variation is indeed there in the final product.