Procedural Spaceship Generator

Hey there.
I've been working for the past few weeks on a procedural model generator in Unity. The idea is to combine set pieces using a generation tree and linking them together.

After working on the base set of the script for a while I'm working on implementing an algorithm to generate simple starfighters.

Here's my progress so far:

 

Cranking up the details.

 

This is an asset I did not expect to see. It definitely looks interesting so I'll give it a watch.

 

uhhhh that looks Nice. i keep it in my eyes.

 

one question. they are walkable ? with rooms / corridors? (simple rooms not complex )

 

@Billy4184 and @neoshaman might be interested in seeing this.

I did try something similar once, but it was a lot of work to set up the "spawnpoints" for different parts on other parts. I abandoned it because I thought that a more random process where then afterwards a human picks the few cool looking things from might be more practical for me.
I'm very interested to see where you are going with this!

 

I was already there lol, it's a slot system like no man sky

 

That's a very interesting project @FPires. Would be interested to know how much and what kind of information is preset before the model is generated?

 

Here's a few more demos:

 

As it is right now, not much. It is very verbose, but the parameters are simple. For example, once the script creates the cockpit, it then gets a connector using custom parameters to get a random connector with a "body" tag, and from the back of the cockpit (or more rarely the bottom).

Afterwards it adds the body passing another custom set of parameters, etc. The tool itself is an API to connect these or any other custom blocks.

It was, yeah. I made around 250 models and they have 5 connectors on average. The connectors can be points or meshes so that a greeble can spawn anywhere in the connector surface.

Not really, no. I intend to do this eventually but since this is a personal project I'm doing on my spare time, "eventually" might be anywhere from "one year from now" to "never". It's not a priority because it can be quite challenging to do it right.

 

Some coloring. The mesh vertices have different RGB values that are used as a weight for the shader. The shader can then selectively color the mesh based on these weights.

 

This is amazing! :O
Eta?

 

These look pretty nice. But I notice most people who make spaceships or alien ships seem to always make very 'realistic' looking mechanical objects which are a bit lifeless. I want my alien ships to be more organic and alive and less like some kind of machine. Just a thought.

 

I definitely agree with this - I've always seen my spaceships as character-driven, like saddled animals and I've always been impressed with 'organic' hardsurface shapes myself. It's one reason why I've tried not to go down the route of too much greebles myself.

I'd be interested to know OP if you have any plans/ideas for how to achieve a more visceral impression? I think it would be the next step for making your generated ships more appealing. I think it might be interesting to attempt to fit a set of rules for placing parts that corresponds to a human/animal body? Just a thought anyway.

 

My original goal was to make ships that look like the traditional human-made machines in series like BSG and Star Wars. I'll have to clean up this tool a lot before I release it, but If I ever do release it I'll keep your needs in mind.

Right now the tool is divided in four parts:
* The API
* The Addons
* The Mesh blocks
* The Algorithm scripts

The main aspect of the tool is the API to link the blocks together. You can do whatever you want with them via script, so if I want to make a script that places ship parts corresponding to animal parts, that should be very trivial to do.

The blocks themselves are customizable and so are the algorithms, but I do have some ideas that could improve on the appearance of organic ships. Right now I'm working on a system to generate viable materials for the models. Once I'm done with it I'll research my ideas for organic ships.

 

Really happy with this:



The generated model now splits UVs with no overlaps. The usage of space isn't 100% optimal, but the generated UVs are scaled proportionally to the block size relative to the final model, so everything is uniform - a single texture applied to the whole model won't make the small greeble look super detailed compared to the cockpit, etc.

 

In terms of generating a ship, do you have a nice user interface with sliders etc where you can adjust certain aspects of what gets created, then have the system automatically generate the result? And could it generate lots of different variations quickly?

 

that was perfect
something like that need i for my project.

 

Sounds and looks pretty cool!

 

Not right now, no. There's no interface.

I have a roadmap to do right now which I've been progressing at a fairly decent pace. I'm gonna post it as I do it so as not to create expectations. If I can complete the roadmap I'll release the tool, otherwise I'll keep it to my own projects. I say this because part of the roadmap is refactoring a lot of the code so that it is accessible and readable to possible buyers, and fixing a lot of bugs which make it simply unrealistic to release it right now at any price tag. There's 2000 lines of code of linear algebra and it's not uncommon to find an oversight here or there.

I have a very good idea of how I want a possible interface for this tool to look like. I wouldn't release it without a very good and intuitive GUI to use.

In other news, I got some improvements on my UVW Mapper and fixed some bugs.



When dealing with procedural textures the important aspect is to keep things uniform and non-overlapping. A nice improvement would be for it to recognize the empty spaces in the UVW of individual objects and take that in consideration to use the final space better, but I'm very, very happy with where it is right now.

As it is, a ship with 40~50 parts is taking 0.09 - 0.1 second to generate which more than fits my goal of being able to generate them during the loading screen for my game. Once texture generation kicks in this number will probably go up a bit, but I believe it will still very manageable.

 

I cant say this enough, man| This is amazing!

 

This looks really cool, looking forward to see this with textures!

 

Reminds me a lot of http://ship.shapewright.com/ - an inspiration, perhaps?

 

I didn't know that existed. The only procedural model tools I know of are building/city generators.

I can see the similarities, though I imagine they are a bit unavoidable when it comes down to placing blocks together at random. Pretty cool website, his tool generates some very cool looking stuff.

 

I've been playing with texture operations and planet generation lately. I started with real-time noise generation for procedural texturing, implemented a texture operations suit, and then realized I could start playing with quadrilaterized spheres to create planets:

 

Is this project still live? I need it...

 

Really cool. I can see a use for this. Hope you can release it.

 

Hey. I'm still working on it. My time has been mostly focused on my game right now that I finished the tools for my purposes. Once I'm done with my game's code I'll revisit the asset to launch it.

Before launching it I intend to clean up some bugs with symmetries and add some additional features such as themes (being able to replace the models from an existing generated object with equivalent models from other themes, such as a high-poly theme that simply improves the look). I'll make a roadmap when I'm closer to it.

 

If you *coff* *coff* need beta testers *coff* call me

 

Looks stunning!

 

I probably will, though I'm a bit far off.

Thanks!

 

A bit of an eye-candy - more on my quadrilaterized sphere script. Each planet takes about 11-20 seconds.

 

Don't worry, i can wait Although i can help you with some minor optimization if necessary. (In case you haven't noticed, i absolutely love this thing)

 

Any updates?

 

I'm still crying myself to sleep every night because i don't have this tool

 

why cry?, write ur own. =)

 

"Why be poor? Make money!"

 

I don't have any updates yet. I'm super busy between work and writing my game, and time has been extremely scarce even for that. With that said I haven't abandoned the tool or the thread (I'm here, right?), I just don't have any ETA.

Also I've been getting quite a few questions regarding procedural generation. I don't mind answering them, but if you have any question I'd appreciate if they were posted here so the answer could be shared with others.

 

Planning on open sourcing?

 

No, but I don't see any reason not to answer some stuff publicly. For an engine this big there's a surprisingly low amount of resources on some areas of development and it'd help people in the future who are googling for solutions.

 

It's been a few months, and I'm curious to know how the project stands. It's definitely beautiful, and google has it listed in the top five in the search result for "procedural spaceship generation".

Anyways, I'm glad I stumbled across this page.

 

Hey. I haven't abandoned the project. I'm actually working full time on it, albeit in a different manner - I'm using the original algorithm to write a procedural shmup.

Some videos depicting the progress over the past few months:

If you have a SomethingAwful account, I've been posting my progress here. I intend to open up a site/blog in the future when I have some more content.

Albeit it's not user-friendly and with no GUI (only scripting), the asset works for my purposes right now, and working on the asset is secondary to the game as it is. I want to work on my game before I release the model generator.

However, I haven't abandoned the idea of working on an interface and release the tool. Just recently I started refactoring the code in order to add some new features. This gave me the opportunity to clean up some of the code and during the refactoring I coded thinking of user-friendliness. So I'm still developing it while keeping in mind a release in the future.

A lot of features that I believe to be necessary for the asset are also things I'm developing and battle-testing the best way possible, with my own income as a game developer depending on it. A few things to name:

The essentials:
1) Algorithm speed. Trying to optimize to make this as fast as possible.
2) Fixing any memory leakage whatsoever.
3) Bug fixes. When dealing with something as delicate as this, there are so many. Just recently, for example, I fixed an oversight which is what happens when a model crosses the 65k vertices. Since the original ships were fairly low-poly, this took a while to happen.
4) Flexibility. I rewrote a lot of the stuff in my script to make it as generic as possible, without needing to tie it to my own game.
Possible Features:
1) Possibility of multiple algorithms for different objects. I've had this in mind for a few months, so everything I write I write thinking that in the future this program might be sold as an asset to someone who wants to generate a house, a city, or a procedural sword.
2) Simplified collision meshes
3) Support to procedural textures. This is halfway done, and the algorithm allows you to include either existing textures or substances.
4) Support to "skins". Basically, allowing multiple themes to apply over a generated object. This way simple low-poly blocks can be used and substituted by hi-def high poly models, for example.

Anyway, there are other features in the works. Right now I changed the scripts to use a faux DNA system that allows me to mix different algorithms.

I really want to see what happens when I mix an X-Wing with a TIE Fighter.

 

The project looks good!

Edit: The somethingaweful page (looking at it just now) is extremely detailed and helpful. Thank you for all the in-depth notes.

 

After reading through a lot of the SomethingAweful posts, I have to ask how the idea of factions might be approached, while maintaining the idea of procedural ship generation. In other words, is there an optimal way of producing a wide variety of ships, with each ship having a distinguishable "style" or "class" or with your words, "DNA".

What comes to my mind is either assigning a group or class of procedural ships a certain set of parts/polygons unique to that group--or assigning to that group a unique tree of possible orderings of parts/polygons (where the polygons are still available to all groups). Then there's probably a middle ground somewhere.

However the idea is worked out, I look forward to the next SA post.

 

There are model groups (Wings, Body, Cockpits). Some of them are heavily specialized (Wings_FactionX). These contain 20 to 30 models.

There are also collections, to make my life easier. These have weighted values for groups, so a collection of "Assorted Appendages" may have a 30% chance of spawning a wing, 20% chance of spawning an antenna, etc.

There are Ship Algorithms which are pointed from the ship type enum. So the reference for an EmpireFighter will refer to a specific algorithm. These algorithms are split into many functions, unrelated to each other. For example, AddWing() from an EmpireFighter may do something different than AddWing() from a RebelFighter.

The DNA is a sequence of pointers, delegations and variables which define which function from where the ship is gonna use now. So a DNA may be a sequence of, say AddCockpit, AddBody, ExtendBody x4, AddWing, all referencing the EmpireFighter. However you could replace AddWing from EmpireFighter with AddWing from RebelFighter in the instructions, and now you have a DNA with mixed instructions.

I'll detail it further in the thread, but hope it answered your question somewhat.

 

Hey FPires, did you ever finish this procedural ship project? Those models look incredible. If it was on the store today I'd probably buy it.

 

Yes, this is really looking great I'd be interested, too.

 

Abandoned?

 

if you read his last message, he's posting development on his personal forums. he won't be releasing the code.

 

Okay then. I'll make my own
Starting now.