How to make an AAA game in Unity (or fail badly)

@neoshaman I checked that out only yesterday, and have kept an eye on it for a while. It's pretty interesting, but it seems like something that's more suited to pure procedural generation. I admit I don't really know anything about it, I've been more interested in other approaches.

The way I envision what I'm looking for is as a level editor of modular pieces, something that can easily assemble corridors and rooms and swap out the meshes. At this stage, I think the best thing would be for me to make my own stuff, because nothing I've seen would quite do the job and add customization where I need it.

 

What do you mean by "more suited to pure procedural generation"?

I have read the 13 pages before sharing it to see if it has enough versatility. It's not a general modeling tools, at least not yet, But it do allow to make quality adjustable building in 5mn clocked (as seen with beta tester). And while it's easy to waste polygon by going crazy with the node, the author has demonstrated it's flexible enough to allow sharp optimization (also in less time it takes to open unity on my computer). It's more gear toward architectures generation but it's possible to generate prop with it.

It's not a level editor at all, it's not design for that. Though some are already using it for that. It's more of a discrete parametric object generator.

Also a lot of problem of people who like making procedural or parametric tools is they like to emphasize the impossible, scifi and fantastic, when a better demonstration is the historical, the mundane and the familiar (which he also have demonstrate).

But I looked for stuff that closely match what you want. If I'm correct, you want something that turn a "schematic doodle" (high level structure) into a fully developed level in which you can swap elements that don't fit the vision? Ie more like a refiner or the neural doodle I shared earlier but for level design?

 

Well it looks great, and is definitely newsworthy, but may not be what I need. Yes, basically as you said, I'm looking for something that can take high level information and fill in the low-level information. I'm not really sure what the best kind of input information would be, but essentially, since I'm making a sci fi game, I imagine it as something that would be capable of:

• Building a certain type of architecture between two points (editor handles) (i.e. a corridor, stairs, lights, whatever);
• Automatically building volumes e.g. rooms out of modular pieces, where these volumes are easily editable in 'primitive' format;
• Automatically scale the geometry and shift the uvs of meshes a little here and there where necessary (e.g. for a corridor that comes diagonally off a doorway, when the modular pieces were made for only perpendicular ones);
• etc. (can't think too well right now)

The parametric modelling think seems like it may have unforeseen constraints, but again I haven't really looked into it - having only googled it right now, it seems like in theory everything could be described in a parametric model if the vector is long enough! Obviously the author is working on a specific implementation but not really having any more info I'm not sure if I could use it. Definitely an exciting project though!

 

When you describe what you want, it's parametric modeling, It's not a technique per see, I just avoid procedural generation other the drama this name provoke you can also say algorithmic placement or modeling , it just mean you input parameter that control the process toward desired result!

But yeah there is plenty tools that implement subset of what you want (mega shapes, mega scatter, curvy, erc ... on the asset store are generic building blocks) , there won't be a single tools, the auto scale is what archimatix do, you create a blueprint and then generate architecture from it (ie number of floor, nuber of doors, number of polygon, etc ...).

There is no silver bullets, achieving this will need a set of tools that feed each other.

 

OK well as I said I don't really know how the whole thing is being put together. I did notice that a lot of the screenshots showed stuff that appeared to be very 'procedural', i.e. geometrical/repetitive but that might be just for demo. The utility really depends on the user interface and if it can easily do what I want it to.

Let us all know how it goes if you ever use it, and if/how it might help in creating bigger and better games.

 

Okay, just trying to provide information.

On a bigger subject, I'm bit surprise about the geometrical/repetitive statement, because it's a property of "architecture" and generally even nature. Generally two things break that, composition and details, both working hand in hand. Composition by orienting the gaze and building meaning. Details by breaking monotony and adding texture, flavor. Artificial stuff tend to tip heavily toward composition while nature tip toward detail. The thing is procedural generation is good at the two stuff. Generally when people say something is bland, I register that as the art direction is bland.

Which made me thinking, this tool is design to compose artificial structure (ie architecture) why do people register this as "bland" instead of just a tool? We can add details and decorate them appropriately afterward either procedurally (substance) or use composition to place them in a good scene.

So I'm thinking, maybe the problem is composition itself, most procedural generation focus on explicit description of "structure" or "chaos" ie implement rules that describe how the thing is put together. However aesthetics and composition is generally mostly "implied" by the coder who implement the technique (through constraint), rather than describe or generated. What I mean is that we will try to find a good set of constraint that put object on the "rules of third" (implicit) rather that implementing the rule of third directly and let the system generate based on that (explicit).

I wonder if we can't have high level system that checked and generate art direction? Maybe that's what is missing into procedural generation, after all art is made of very explicit set of rules. I'm not saying it will replace intent and generate meaning from life experience, but maybe it can narrow the problem a bit much to avoid bland stuff. Ie to create context.

For example no man sky didn't made anything new, space exploration game with crafting, fighting, looting, exploring exist since a long time and have a long history with procedural generation. It seems what differentiated them is that they added explicit art direction generation, for example the artist added code so that colors were always pleasing in relation to each other, they implemented color theory in the algorithm.

I'll try to dig in that direction.

 

Was messing around with active ragdolling. It took a lot to really figure out how to make use of it - my case is not really physics driven, it's just using the physics engine for some additional animation not for impact detection.

Result ain't bad.

 

Well animation is going down with neural networks, I wonder if there is enough data on the asset store and some can potentially be generated by blending. Why not just blend directly? well if it create a generalized model it might generate animation just like we want given selected parameter.

 

This is 5 years minimum from being directly useful.

But it is pretty friggin cool!

 

@Billy4184: Just found this:

https://github.com/a1studmuffin/SpaceshipGenerator

 

Looks pretty cool, thanks for the link! As an end product I think it still leaves a bit to be desired aesthetically but looks like a good starting point, especially for 'mammoth' ships and stations. As bad as it sounds (!) I'm leaning more and more toward procedural generation as a foundation for my game, not that I think it will be any easier but probably more practical for a solo dev ... really depends on how well some experiments turn out that I have in mind.

How's your animation project coming along?

 

Not much. Severe lack of motivation and recently also lack of time.

 

@neoshaman:
https://www.artstation.com/artwork/dgkv1
This may or may not be interesting for you, because of the curly hair on the model.

 

Building A Galaxy: Procedural Generation in No Man’s Sky

 

I think this new DAZ HD style isn't very realistic. The pores and scars are exaggerated to a degree that seems to say 'notice this' to draw attention to how realistic it looks, but it doesn't look more realistic, just more incorrectly detailed. To me, some cartoon styles are more realistic because they exaggerate the personality of the character and that's the important part of the interaction with the game.

 

Hearing DAZ mentioned in relation to Uncharted 4 is quite puzzling to me, but OK. For me the DAZ stuff always was associated with a plastic-like "too perfect" look, but I don't follow what they do, so it might have changed.

I get what you mean, although it doesn't bother me. I could imagine it's one of the compromises they need to make in regards to consoles. After everything is compressed down and rendered at 720p, or whatever they now manage to get out of the ps4, many details that are more subtle would just be lost I guess. You often can see this on fabric patterns too, they have to make them a lot more coarse/upscaled than they really are, so that the normal map still can reasonably show the pattern.

And then there's the philosophy of "imperfection is the digital perfection". Stuff being too perfect is a telltale sign of CGI.

What you refer to with the cartoon styles probably is related to the uncanny valley:
https://en.wikipedia.org/wiki/Uncanny_valley

To me cartoonish styles usually make immersion a lot harder, but I'd chalk that up entirely to personal preference.

 

You must not know much about DAZ but they are top of the line modelers and that's the too sweatly, too glossy, too scarred, too big of skin pore style DAZ has with their HD models is the 1st thing I though of when I saw that model on that page linked. Of course with better HW now I expect to big business gaming world to catch up with where DAZ & Poser have been for a long time.

 

Crossing it ? you will need to hire actors to make your game.

 

Super thanks!

Although she is the stereotypical hollywood black girl, ie brown with curly hair, I'm shooting more for ebony (ie darkest) with coarse hair ala "lupita" ... but even her get photography "lighten up" so you are never sure with reference. But more like this, which is a challenge of its own:



Notice how the hair have super tight curl (maybe 4mm of diameter), it's call the 4C type of curl.


I'm really shooting for the extreme, ie 4C hair

Black hair is itself very complex:

Spoiler: More 4C black hair

One thing is that when looking for reference, it's mostly amateur photography with poor lighting. Professional tend to either straighten the hair so much, or just flatten it into a dense silouhette. Finding idealized and professional photography (like for straight hair) is extremely rare and even then there is a clear lack of language around expressing and valorizing this type of hair.

Black hair is next gen state of the art

That said, I'm still struggling with shader and modeling experiment because I need to make anisotropic image to model curl direction, there is no soft to bake, generate or paint anistropic map So even the most naive test I can't do right now. I need it to experiment with specular lighting

 

That's actually cheaper because it's far less labor intensive and far more accurate so it's just business sense to use less time, less modeling, and better animation captures to make a game.

It's getting to the point that you think, hmmm, video camera, motion capture, and digital camera and you're set. Who needs all that 3D modeling school pizazz and all that time & expense...

For some games, like those WWF Wrestling Comedies you can see the game players going all in to have realistic representations of themselves in the game for a hoot.

 

LOL, if they could do 3D models that looked like those ladies, real actors would be in big trouble.

 

They don't have to look like that lol, it's just the approximation of hair and skin I'm looking for.

Spoiler: Good image

That said Mafia 3 is very promising to look at, their character isn't state of the art, but the main is black, have hair with different textures through haircuts, in fact it has better hair rendering than the white character such a rare feat

Spoiler: mafia 3

Final fantasy 13 sazh and dazh are great too! They use similar technique used for tree rendering.

Spoiler: sazh and dazh

I have rarely seen as good rendering, even in actual CGI character

Mgs5 also have some good afro hair rendering ...

 

Since you asked:

This is where I'm at. Still no gameplay.

But I have a question. I'm experimenting with making the enemy tanks rigidbodies and driving them with forces. So far I'm not finding any settings that don't lead to ridiculous behaviour. I'm looking for something lightweight (needs to support 100+ units at the same time - so no realistic tank simulator setups) but realistic enough to get some nice physics interaction going. I've made a few attempts at googling for a solution, but I don't know really what this would be called. Any suggestions would be appreciated.

 

@Martin_H I've never really done car/tank physics before, but off the top of my head it seems like the best optimization would be to avoid wheel physics, i.e., make it basically a hovercraft that happens to sit level with the ground. That way you're just dealing with a single set of xyz forces and torques.

Kind of depends what physics interactions you want.

PS Looking very good there! I don't see anywhere near enough strategy games being done (if that's what it is?).

 

Dude!! That looks sick! I donno what you intend to do with that. But hot damn it looks good. The patterns on the mountains and snow are awesome too.

That game just looks like something that needs to be played. Makes me think of a godzilla style game. You run this crazy mech thing on an insane rampage while the pititful army with their wimpy tanks try to stop you.

Also you gotta tilt shift that!

 

Did somebody say strategy? The AAA studios of the world aren't going to tremble in awe, but here's a shot of Black Hand, my implementation of Diplomacy (caveat: placeholder graphics and models everywhere; just look at those ugly ocean borders...). Dip is a turn-based strategy board game, as most of you know. "Black Hand" is a reference to the Austrian assassins who arguably kicked off World War I. I'm doing a sort of 2.5D map view which you see below, and the off-map interface (e.g. not issuing orders or reviewing results) uses a full 3D room -- basically your diplomatic offices, matched to whatever world power you're playing as (Turkish, Russian, French, etc.) ... so my wife has a whole bunch of artwork ahead of her. No point in showing that, it's all white-box placeholders.

I'm about a month and a half into this. The SQL server, database, data engine, IIS, and REST schema and server are all up and running and being expanded as I add to the client. That all took about 10 days to frame out and test (my corporate-cube-farm programming background really came in handy there). The hard part of a Diplomacy implementation, the "adjudicator" that processes player moves, was already written from my never-released XNA version. That took about four weeks to write back around 2011 or so (massively recursive, very complicated), but only about one day to tweak for this new project. The client is mostly stubbed out, I was focusing on a single path to get to the point I could work on gameplay, but using real data -- so the basics are working: login, selecting a game session, loading game data, populating the map, and now I'm working on order entry. Probably three weeks of work in the client including about a week in the pretty complicated (for me) shader that handles most of the map FX (border highlighting, region coloration, the water animation I'm too lazy to video right now, etc). Two weeks here and there building (and using) a map editor. I put a lot of time into that since I hope to support multiple maps (sort of a Diplomacy holy grail) and the map data for the standard Western European Dip map is nearly 1,000 lines of data (region adjacency, model positions, game setup information, etc.) and since the map coords are normalized it's not something I want to write by hand.

I know board games won't set the world on fire but turn-based strategy is kind of a passion of mine and something that doesn't get much attention, and Diplomacy in particular is something I've played on and off my whole life as time (and willing friends) permits. Everybody has to start somewhere, right? I figure I've done enough yapping here, it was high time to throw out a screen shot and prove I'm Actually Doing Something.

P.S. Martin_H's tanks look waaaay better. lol

 

@MV10 Great stuff! Very nice to see more examples of work.

 

Aw, thank you guys! Your feedback is really encouraging!

That's pretty easy to get with Worldmachine. Perhaps you see now why I don't want to use the Unity terrain tools at all. The setup I made for that allows me to feed it a greyscale image with very roughly painted in height values (basically just 3 different heights of playable area, smudging their transitions where I want a walkable slope to be created, and an impassable outer "wall") and it generats this terrain for me. I haven't even added a fancy terrain shader there yet.
This was from another test, but I've since messed up that particular setup, trying to "improve" it:


If you want something similar you need a terrain shader that can blend a colormap across the whole terrain, a Worldmachine license, and about a week or so of time, at the max. I consider WM to be an easy to use tool, it gives relatively quick feedback on your changes in the lowres preview, and for getting something like what you see here you might be able to just feed your heightmap into an erosion node, use the basic coverage macro that comes with WM and tweak the colors. I'm guessing you don't want to change your heightmaps, since that would likely interfere with what you've build already, but you might get some use out of just the full terrain colormap with some visible erosion. I have to warn you though, that it can be quite RAM-hungry. I've got 64GB in my system and I've once seen WM basically complain "that's not enough RAM, are you really sure you want me to swap to disk space for this simulation?". I think tiled builds have solved that, but I don't want to blindly recommend you a tool that might not run on your system.

http://www.world-machine.com/download.php

Seems like they have a trial version. You might need the pro version to really get full use out of the tool.

I don't quite know what to call it. The gameplay I'm going for is:
- you've got this single giant unit, fighting versus an army of small units
- on your titan multiple cannons are mounted, like on a battleship
- the turrets all have limited firing arcs
- the torso of the titan does not turn, you have to turn the whole thing to manage firing arcs
- you can walk independantly of the rotation (so you can micro it to circlestrafe for example)
- the hull can tank massive damage before it's destroyed
- turrets can get damaged easily but can be repaired again
- you need to man turrets with available crew members for them to be functional
- if a turret gets damaged while manned, the crew dies
- idle crew not manning or repairing a turret is safe deep inside the hull of the titan
- when you've only got 1 guy left you can still move, and selfdestruct with a massive explosion, to retain a bit of agency till the bitter end and make the emergent narrative of your playthrough attempt that tiny bit more interesting
- you can move the camera independantly from the titan (maybe up till a certain distance)
- different turrets get different keys assigned, hold key = aim towards cursor, hold key + left click = fire turret
- turrets have different speeds at which they adjust their aim
- meta progression is unlocking new titan configurations

That's what I've had to come up with, to make myself care a bit again (remember my "making people care" thread?). The twinstick-ish concept I had before wasn't cutting it anymore.
I currently have no plan of a commercial release, so I can make it as weird and unique as I want and the target specs are my own hardware (i7, gtx 670).

Long term it needs to be possible to hook it up to an AI with pathfinding. Short term I'm looking for driving forward and turning at a fixed rate, without it looking like crap or causing issues. I'm using convex mesh colliders at the moment. One of the many problems I've seen so far is that when I move them with forces, the friction sometimes locks them in this position while driving:



But if I lower friction too much they slide like on ice (I'm going for desert by the way, not snow). If I just use Rigidbody.MovePosition in the forward direction on every frame, they start flying away on any bump. If I use MovePosition only when they have contact to the ground, they would drive up a ramp and then abruptly stop mid air because that doesn't give them momentum. Also if I use MovePosition and some other Rigidbody interacts with them, I'd imagine it to act weird too. I've tried moving the center of mass below the ground, but then it looks super weird when they fly.

I want to end up with a practical solution that does not produce any situations where a player would go "that looks weird". Despite the pixel look I want this to look as realistic as feasible. E.g. the feet of the titan still slide and clip 2-3 pixels per step. That's not acceptable. They need to be fixed on the ground, or the whole perception of the thing as solid, heavy, and steadfast breaks. These are things that I care a great deal about and see devs make huge compromises on, time and time again, even in AAA. I'm willing to prioritize such details over most other aspects.

 

@Martin_H, how about 'hovercrafting' it, i.e., raycasting down, adjusting the height with forces so that it floats above the ground, but drop the mesh until it looks like it touches. From above this should look pretty good. The easiest way to avoid issues with the ground is to not be on it in the first place! A PID controller (which is a lot simpler than it sounds) is very good for preventing overshoot and giving the impression of friction.

If you want some physics interactions with the ground, definitely put some wheel colliders on it (4 should do) - I think this is pretty standard for tanks in relatively simplified settings. Definitely don't just slide it around over the ground, especially if your collider is a box. I wouldn't be surprised to see a lot of the sticky scale issues (i.e. sticking and wierd stuff when two objects are at different scales) with terrains although I pretty much stick to space games!

My first ever game project had a tank, it was a 2D game (3D with ortho camera), and I used wheel colliders, worked for me.

 

I always wondered how to get that look. Actually a bunch of my first posts here was trying to get tips on how to do region highlighting and border stuff.

More recently I also wanted to do los indicators. Seems like a ton of work that stuff. OTOH, once you do it - the results tend to look friggin awesome.

 

If you mean border highlights, it's actually pretty easy. And I say that as a total shader n00b -- always knew the theory but this was my first shader.

On my first attempt I had the shader figuring it out on the fly, but it was far too expensive. The shader has a base texture which is most of the graphics you see, a region mask which uses the red channel to indicate the region index number (both in terms of map data and other lookups), and a border mask. (There is also a water texture and water mask and some alpha sliders.)

Before I added the border mask, the shader samples neighboring texels to determine if a border was nearby and calculated an alpha by adding the number of samples belonging to a different region. But a 10x10 scan is 121 samples so that added over half a billion extra samples to every pass on my 2K texture... and that created a border only half as wide as what is in the screen shot. Sub-optimal, as they say. So fudging it with an additional border mask made it fast and easy. A small copy of the region and border masks are shown below (the border mask is a cut-out around Denmark so you can see the effect; shrinking it down made it hard to see what was going on).



The border mask is just Photo Shop's stroke and inner glow. The region index from the region mask red channel is used as a lookup against another 128 x 2 texture created at runtime -- the first row is the color to apply to each region and the second row uses the alpha channel to toggle either full-fill or border-only on the region.

Later I will improve memory and performance by combining all these masks (so regions in red, borders in green, and water in blue) but I was getting burned out Photo Shopping and not coding, so it went on the to-do list.

I guess maybe it sounds complicated, but in reality the code is remarkably simple stuff.

Getting my fancy ocean water moving without being distracting -- that was complicated. Around 130 lines of code for water, versus 10 lines of code for all the stuff I described above.

 

In fact, managing all the layers in Photo Shop to produce those masks was more complicated than writing the shader. (I really, really dislike Adobe...) That one PSD is 90 GB...

 

It wasn't an Austrian assassin but a Serb national who was a subject of the Astro-Hungarian Empire. Good diplomacy could be argued as refusing to teach the past political histories when all the participants are dead and unpunishable especially when the perspective of who was right and who was wrong is subjective.

 

The really stupid part is that I've been naming my Diplomacy-like games "Black Hand" since the 90s and I explain the origin of the name incorrectly Every. Single. Time. That's what I get for poking my head up out of the OCD-code-hole and getting sucked back into this thread.

 

I wouldn't worry too much about it, but as with the time everybody but one person in my Logic Design class missed a question on an exam because the Logic Design book gave an example that was wrong and they memorized it, that doesn't change that the book and their answer was wrong.

Speaking of being unproductively drawn back into the forums I (try to) make my exit for the next month or so now from the forums too...I know what you mean there. Ciao

 

What resolution and bit depth is your document? I've worked on files with over 800 layers/smart objects, but never have I managed to get a file to 90 gigs on the hard drive. The scratchdisk swap files can easily get that big though.

First tests indicate I'm on the right track with that approach, thanks a lot!

 

This looks incredibly interesting. Not just graphically (I do think it looks awesome very clean, readable, nice colors and the designs of the giant mech and tanks are great too) it looks like an interesting scenario as well. I agree with @frosted it reminds me of watching the old Godzilla and other monster movies. Great stuff! Looking forward to seeing some gameplay. Always the most important part!

 

It really does @Martin_H, there's just some quality to it that makes you want to play it. I think this is a really rare quality in a screenshot, and very important. It just looks like fun. Physics will need to play a big role in this.

If you can make the gameplay half way live up to that image you will have something pretty awesome

 

I'm very happy and a bit surprised about the amount of positive feedback I'm getting =). Considering that objectively this is just a screenshot of 3 assets (mech, tank, landscape). This doesn't qualify as a game yet by any stretch of the imagination ^^.

But it certainly is a good sign, that you already can start to feel what I'm going for here, and you've got me motivated to push forward with this!

 

The right 3 assets in the right setup. Makes all the difference in the world.

Seriously. Tilt shift. You can get away with way more wonton destruction with a tilt shift mood, and it will totally fit.

 

I'm wondering what exactly you mean with that. Do you have an example? When I hear tilt shift I think of this:



It's a technique in photography that can be used to make big things look like a tiny model train landscape. As you can imagine, that's the opposite of what I'm going for. I'm going for a kind of aerial drone view, and that doesn't have DOF at all, like you can see here for example.

 

I think the toy feeling could really hit the right coord, it could also help give you space for more wanton destruction of civilian cities and stuff.

 

Drone view...
Have you ever played a flightsim?
Basically, in this kind of scenario you can use one polygon for the whole ground, put a photographic texture (it can even be unlit) on it, and it will look great, as long as you don't let the player get too close to the ground. Check out il-2 sturmovik and Microsoft Flight Simulator.

I would advise to take a clipstudio a try. It is fairly cheap.
It also has interesting ability where you can put blending mode and effect (such as recoloring) onto layer folder (instead of applying them to layer). So the folder first blends its contents together and then result is fed through the blending mode of the folder. This way you could, for example, easily create splat maps from original image. And I'm quite sure it won't result in 90GB psd file.

 

Thanks for the suggestion, but for this game its not gonna happen. I did try that sort of effect with the Scion DOF while I was playing around with its settings, but it just doesn't fit my vision for the game. For e.g. a 2.5 D RTS where you control ants on a kitchen counter, and want to blur off the unwalkable areas in front and back, it would be perfect. Here I'm rather going for a gritty and bleak "last stand" kind of fight. Sorry if that disappoints you ^^.

I played War Thunder. But I had a perspective a lot closer than a flight sim in mind. Think more Parrot AR drone than Predator drone. The screenshot of the mech that I posted, is already pretty close to the camera distance that I'll likely will end up with. But Thanks for the suggestion.

Performance isn't a big concern, since I mostly work on this for myself as a hobby. My GTX 670 seems to have died today. I'll likely get a GTX 1060 once it's released. CPU already seems to be the bottleneck and that will only get more pronounced once I get a stronger GPU.

That's a very useful feature. Photoshop can do things like applying layestyles to layer groups too, since CS6.

 

I second that... used something similar for my first rough Character Controller prototype... easy to do, and with a single short raycast per frame not so expensive.

To make it look like real physics are at work, I am now modifying it slightly... Don't know if at one point it start becoming inefficient and a rigidbody/physics setup would become more performant... but at least I will not have to fight with PhysX as much... had enough of that when I tried to build an offroad racing game with wheel colliders.

What I already have implemented is adjusting the speed of the Unit by looking at the normal of the terrainCollider. When raycasting, you get back the hitinfo... that hitinfo struct also contains the normal of the collider face the raycast has hit, which is extremly handy for a lot of effects (used it extensively for my armour model with my last prototype)...
in this particular case I compare the normal of the ground with the forward vector of the character (not the up vector, as Vector3.Angle is "Abs"-ing the result), adjust for the 90° angle between them, and I set the maxSpeed and the acceleration from that.
Add some momentum var so the momentum builds up over multiple frame and needs to be reduced over multiple frames, to simulate acceleration and braking, and you got an actually quite realistic looking "physical movement behaviour" for your units.

Next I will work on slipping on steep slopes, and adjusting the speed and slipping for different types of ground textures.


Really don't know if this is any cheaper than letting PhysX handle it. There are some calculations involved. But then I had to run some calculations just to get PhysX do my bidding every frame (using corrective forces and whatnot), so I guess it will even out.


EDIT:

Oh, and in your case you can use the terrainCollider normal to align your tanks with the terrain.

 

That's very clever, I'll try that. Thanks!

From the last tests I did I'd say wheel colliders seem to work fine for what I want, but I still need to implement the right way to controll their forces and steering settings. I can't complain about performance, I can spawn 100 tanks with 4 wheels each.

I anticipate I'll have to do that too, unrelated to the driving. When the mech stomps on things either they get trapped between two colliders and jitter around, or they are catapulted away with too much speed. I'll try and just cap the speed for certain types of objects and see if that helps, even though it seems like a weird solution. I tried setting different values for maxDepenetrationVelocity, but somehow that didn't yield the results I was hoping for.

But first I'll try and get an even clearer vision on the look that I'm going for, because it'll bother and distract me until I've figured that out. ^^

 

Sweet, thanks. I was lamenting the lack of that exact capability over the weekend. Hopefully it works with CS5, I'm stubbornly refusing to play the subscription game after spending vast amounts of money over the years buying Adobe products.

And it's 9GB, I typo'd 90. Still a bunch o' pixels though.

 

I arrived at the same technique in my first-ever Unity project, this goofy little screen saver. I had problems with the raycast occasionally returning very crazy normals. I lost interest before I figured out why. I worked around it by averaging multiple raycasts since I just had one moving object.

But more relevantly to the discussion, I found it useful to add forward-aimed raycasts to "predict" the terrain ahead. That allowed me to begin tilting back or forward to go up or down the next hill, rather than waiting until I nearly ran into it. Then I cast left and right (but also ahead a little) to smooth out the roll adjustments. The end result was a pretty good hover effect.

 

Well, keep in mind my expierience is mainly with Unity 4 and PhysX 2.8 ... when Unity 5 broke my physics setup (which relied on heavely) and game code, I kinda gave up on my prototype and shelved it.

When I tried the new wheel colliders they looked even more broken to me than the old ones. The old ones (Physx 2.8/Unity 4) where working fine, just not... really realistic. They were ommitting a lot of behaviour that was needed for the car to "look realistic". Like loosing grip on slopes. Reacting to the ground as if it was the actual medium it should represent, and not hard metal (which led to a very bouncy offroad ride you would never get when driving over sand dunes for example).
And that is before I had to go in and hack Physical behaviour to make the whole thing playable (like adding a stabilizing "gyro" that would prevent the car from landing on its roof...)

Either the new Wheel colliders had all the settings set up differently (different scale or whatevet), or something was broken. When I set up my cars in Unity 5 with the EXACT same config, the shock absorbers would just sag through, the acceleration would be totally off and so on.
Given the amount of complaints by other devs, I guess it wasn't just me.


Now, if that all really matters to you if you have a simpler setup and goal, IDK. But given you try to develop an RTS, I think 4 wheelcolliders per tank are total overkill. A simpler solution like a single wheel collider (stabilized with additional forces), or just plain DIY system like the one I described up there would certainly be more scalable.

PhysX 3.3 might be fast, but your CPU time is limited (especially when the game is played on a slow mobile chip)... you rather waste that time on 400 wheel colliders updating, or more complex AI?

Of course you should do some performance testing on that instead of trusting my words alone. I didn't test this in your use case, and with an older version of PhysX.

 

2 peoples on unreal engine
https://forums.unrealengine.com/showthread.php?109788


more breakdown on the forum

The unreal pro say it looks like the work of 16 peoples (at 6:13)

Though it's artist, I see nothing about game system and scope ... and how long did that takes