Intergalactic Travel

[misterslime]

Will KSP 2 eventually have other galaxies to explore?

[Davi SDF]

I doubt, there are people that have over 5k Hours in the original KSP and haven't explored the whole system, now imagine this but like, what... 500 systems? (for a whole galaxy? even scaled down they would need a lot!)

I think it might be a little too much, and could encourage an "build a base, get resources, build even larger ship, move on" playstyle, wich is kinda... sad? in a way? I mean, you would just forget the last planet because there's so many!

There's also the technological barrier, but i ain't gonna go into details on that.

[SpaceFace545]

This should speak for itself

[K^2]

1 hour ago, Davi SDF said:

500 systems?

Even at Kerbal Scale, you have to keep in mind that this is in 3D, so 500 systems would be a pitiful likeness of a galaxy. You might be able to get away with 3 digit numbers for some of the globular clusters, but for something that will just feel like a spiral galaxy at scale you have to be operating on 5-6 digit numbers at least. In contrast to the nine digit number of stars in our own galaxy.

Now, naturally, if you were to go that route, you wouldn't be expected to visit even a meaningful fraction of the worlds, which means you can get away with considerable repetition, but at least everything a typical player visits needs to feel fresh. So even if that's just a dozen or so systems  on average, to account for birthday paradox, you probably need something on the order of 100 hand-crafted systems which you can then vary a bit with procedural work so that everything feels sufficiently unique. Needless to say, that's a colossal amount of work. Maybe if Rockstar takes over KSP3... 

Another route is go full procedural on everything and hope you make it good enough. No Man's Sky ended up ok. Eventually. But it was no small effort and was designed to be procedural from the ground up. Plus, the creators were fine with it being very sci-fi, so some of it looking... unnatural was par for the course. KSP is a bit more grounded than that, which means that generating believable and yet diverse worlds would be even harder. I don't think Intercept has nearly enough resources for this either.

Now, I will say that I would be very interested to see a modding attempt to build a procedural galaxy. With expectations for quality not being quite the same and some amount of repetitiveness being far more acceptable for a fan project, a talented group might be able to put together something that would be interesting. I don't think it'd overshadow stock systems as The Way To Play KSP2, but it might be good enough to sink some dozens of hours exploring all kinds of procedural star systems and worlds. Could be fun. But this is not for vanilla game.

If somebody will want to give it a shot, I can pitch you some new ideas from the land of Deep Learning used to create believable landscapes. With a decent underlying generator for base relief of the planet based on classical methods that build up on Perlin noise, a DL algorithm can turn it into weathered mountains, river valleys, place biomes, and so on to form a fairly natural-looking world. I wouldn't use this for a gritty realistic game, but for a KSP-style planet, I think it can cut mustard.

[TLTay]

How many lifetimes are you going to be playing KSP2? A couple thousand consecutively?

[kerbiloid]

Quote

400 billion star systems

 

There are 400 billion star systems in Elite Dangerous.

 

[pandaman]

I don't really see the point of adding an 'intergalactic' aspect.  Yes it sound cool, but in practice it would essentially be the same as interstellar, but it just takes longer - much longer.

More than enough variety can be provided by representing the 'local' part of the galaxy that Kerbin is in.  'Local' here still being  a very long way, but quite close relative to the rest of the galaxy.

[mcwaffles2003]

7 hours ago, K^2 said:

Now, I will say that I would be very interested to see a modding attempt to build a procedural galaxy. With expectations for quality not being quite the same and some amount of repetitiveness being far more acceptable for a fan project, a talented group might be able to put together something that would be interesting. I don't think it'd overshadow stock systems as The Way To Play KSP2, but it might be good enough to sink some dozens of hours exploring all kinds of procedural star systems and worlds. Could be fun. But this is not for vanilla game.

If somebody will want to give it a shot, I can pitch you some new ideas from the land of Deep Learning used to create believable landscapes. With a decent underlying generator for base relief of the planet based on classical methods that build up on Perlin noise, a DL algorithm can turn it into weathered mountains, river valleys, place biomes, and so on to form a fairly natural-looking world. I wouldn't use this for a gritty realistic game, but for a KSP-style planet, I think it can cut mustard.

Have you seen the mod To Boldly Go?

Spoiler

 

 

 

I've been hoping for a hand crafted stellar neighborhood so everyone can share a similar tailored experience but having that stellar neighborhood surrounded by a set of procedural systems just to press out the limits of what can be done.

Not everyone touches the game border but just seeing it can be irritating, while making it so distant it wont realistically be reached can make the world you're in that much more believable.

Edited June 20, 2021 by mcwaffles2003
removed a redundant word

[K^2]

1 hour ago, mcwaffles2003 said:

Have you seen the mod To Boldly Go?

I haven't, but yeah, that's roughly what I was picturing. Native support for multiple star systems should improve some of that dramatically. My biggest complaint is that surfaces of planets/moons look exactly what I'd picture procedural planets/moons to look like. If you are just taking random noise, filtering it down, and then peppering some features, which is all it seems to do, then you only have two possible outcomes. Either bland and featureless or noisy and cluttered.

In the real world, there is structure to features. Terrain tends to be fractal-like, but details change with scale. You can't capture that with a simple generator. You have to do something extra. Some terrain generation tools try to fix this by simulating weathering effects from water and wind erosion and sometimes even some tectonics. But that's extremely computationally expensive and takes time to process even on small regions, let alone trying to replicate billions of years of changes on an entire planet.

But that's where generative adversarial networks come in. There's a pretty good Two Minute Papers episode on one such example. It is possible to train the network so that instead of topographical guide, it would use output of a noise-based generator and spit out a complete elevation map that has erosion features, valleys, etc. It's a bit of work, but most of the heavy lifting has already been done in research and it's just a matter of getting enough training data.

 

[mcwaffles2003]

@K^2

That's a great video and I'm surprised I've never come across this YT channel, thank you.

7 minutes ago, K^2 said:

I haven't, but yeah, that's roughly what I was picturing. Native support for multiple star systems should improve some of that dramatically. My biggest complaint is that surfaces of planets/moons look exactly what I'd picture procedural planets/moons to look like. If you are just taking random noise, filtering it down, and then peppering some features, which is all it seems to do, then you only have two possible outcomes. Either bland and featureless or noisy and cluttered.

In the real world, there is structure to features. Terrain tends to be fractal-like, but details change with scale. You can't capture that with a simple generator. You have to do something extra. Some terrain generation tools try to fix this by simulating weathering effects from water and wind erosion and sometimes even some tectonics. But that's extremely computationally expensive and takes time to process even on small regions, let alone trying to replicate billions of years of changes on an entire planet.

But that's where generative adversarial networks come in. There's a pretty good Two Minute Papers episode on one such example. It is possible to train the network so that instead of topographical guide, it would use output of a noise-based generator and spit out a complete elevation map that has erosion features, valleys, etc. It's a bit of work, but most of the heavy lifting has already been done in research and it's just a matter of getting enough training data.

I get the feeling that this would become very memory intensive if all the terrain needs to remain as is after has been generated. I assume this because I have the impression that unlike procedural systems working off a seed, DNN AI can generate inconsistent results where the next time the same region is rendered if it weren't saved or stored in memory there may be slight differences in the generation. Is this a fair assumption or is it the case that after DNN AI training is paused  they perform perfectly deterministically?

[K^2]

38 minutes ago, mcwaffles2003 said:

or is it the case that after DNN AI training is paused  they perform perfectly deterministically?

That. There are different kinds of neural nets, of course, but as far as pretty much anything related to Deep Learning goes, you have two distinct processes. You train the network first, which is slow and not necessarily deterministic. Once you are done with training, you use the network for evaluation. At that point, the weights are fixed and the process is fully deterministic. Generally, once you released a version of your product or feature, it will only be running in evaluation mode with all of the weights you generated during training supplied. That way, the neural net will produce the same output for all the users consistently.

So yeah, as long as you use the same seed to generate the input, it should always produce the same output. The only place where you have to be careful is the edges of terrain tiles, where you have to make sure to provide a little bit of additional input data from neighboring tiles. But that's standard procedure for this kind of work.

[The Aziz]

Just give modders some time after the game and modding API are out. Sure they won't just port old planetary mods because they'll look bland compared to new and remade planets, but once that's done, and solar systems are figured out, someone, somewhere will make a plugin to connect all system mods and make a galaxy or two out of them.

[Bej Kerman]

2 hours ago, K^2 said:

...So yeah, as long as you use the same seed to generate the input, it should always produce the same output. The only place where you have to be careful is the edges of terrain tiles, where you have to make sure to provide a little bit of additional input data from neighboring tiles. But that's standard procedure for this kind of work.

Each one of your responses have actually been very insightful on how this could be achieved - interesting stuff!

[ballisticfox0]

14 hours ago, misterslime said:

Will KSP 2 eventually have other galaxies to explore?

There are enough stars in the Milky Way that every person on earth could have 50 star systems for themselves

if you saw one star system per second it would take you over 12k years to see them all

and now you want to go to another?

Will we see interstellar travel outside of our local star bubble is a much more reasonable question 

I would like to see procedurally generates systems past the 100ly bubble but to go past 2000ly is just a waste at that point

[dave1904]

Spoiler

 

Not really. Even people that know how far galaxys are from eachother cannot really comprehend the distance.

[Zaffre]

At best we're getting a local cluster of stars. Maybe a tiny "galaxy" like what the old Kerbal Star Systems mod did, but I'm not banking on that. Galaxies are titanic, and with the named planets we've got, they won't be procedural generated like in Elite: Dangerous. No need for them to be. Some extra star systems on top of existing gameplay will be quite enough to play with.

[asap1]

it could be cool but would be kinda impractical as many of us would never reach beyond the local system I personally have never been to laythe, tylo, vall bob or any other joolian moons. same thing with eloo, dres, gilly, and moho. I instead prefer to design crazy aircraft on kerbin test and perfect them and then forget and never use them(guess I am rnd addict starting to aproach the 1000 hour mark). I believe this is partially because of lack of any real purpose for going to other planets( I did go to eve once but once I got there it was a buggy mess very prone to kraken attacks but it was fun)    

[Bej Kerman]

2 hours ago, ballisticfox0 said:

There are enough stars in the Milky Way that every person on earth could have 50 star systems for themselves

if you saw one star system per second it would take you over 12k years to see them all

and now you want to go to another?

Will we see interstellar travel outside of our local star bubble is a much more reasonable question 

I would like to see procedurally generates systems past the 100ly bubble but to go past 2000ly is just a waste at that point

As someone said earlier, it's about making the limits of the ingame world invisible. 

[Starhelperdude]

believe me, even near stars to kerbal are enough, even a whole kerbal galaxy would be wayyyy too big for gameplay porpuses

 

[cubinator]

Maybe the Kerbal galaxy could have a satellite galaxy like the LMC, but I doubt that there will actually be more than a handful of actual star systems in the game.

[Starhelperdude]

Just now, cubinator said:

Maybe the Kerbal galaxy could have a satellite galaxy like the LMC, but I doubt that there will actually be more than a handful of actual star systems in the game.

we just don't need another galaxy, there are plenty of stars already, even in a small galaxy, and the stars wouldn't be different in a seperate galaxy

 

[cubinator]

16 minutes ago, Starhelperdude said:

we just don't need another galaxy, there are plenty of stars already, even in a small galaxy, and the stars wouldn't be different in a seperate galaxy

 

They would be in a different galaxy though! That's cool enough on its own that many players would want to try it.

[Starhelperdude]

Just now, cubinator said:

They would be in a different galaxy though! That's cool enough on its own that many players would want to try it.

you would need soo much delta-v that you would either need pretty much sci-fi tech or soo long fuel tanks that it lags

and other than seeing a galaxy from outside the galaxy nothing would be new

[Guest]

I think it's ok to have other galaxies. Definitely not travelable, but would be cool to point telescopes at.

[Starhelperdude]

12 minutes ago, The Doodling Astronaut said:

I think it's ok to have other galaxies. Definitely not travelable, but would be cool to point telescopes at.

of course, I would absolutely love that, since I like astronomy