Doing It Wrong

Doing it wrong! The story of my life. An act in 3 parts - hopefully ...

I had forgotten how to make characters.

How did I make this? Absolutely no idea.

I hadn't written anything down. I knew the software involved but the actual steps? No idea. So I started again from scratch, intending to make a better system and this time wrote everything down.

I ended up writing a long thread on how I got a model from makeHuman, into Blender, and ready to import it into open source game engine Torque3D. You can veiw that here.

Now importing a mesh into Blender tends to come with a whole lot of transforms which Torque just doesn't like. The main being facing the wrong direction as in-game, negative Y is forward. So for the last nth years I have been seperating meshes and armatures and editing them so they face the correct way and then aligning them manually. This gets rid of all the wonky transforms and sets everything to a nice default 0, 0, 0. That way I can - hopefully - put any other animation into the armature.

This is all very faffy.

Whilst changing the default pose by using the "apply visual pose as rest pose" I was struck by a sudden thought; wouldn't it be nice if object mode had an apply transforms option? Thus I could pose the model with the correct facings and just apply  each object, armature, mesh, etc with a click to default it all to zero. And guess what? It does have that button already.

Well ... crap. I've been causing myself unfathomable amounts of problems for god knows how many years by manually rotating everything by selecting all the vertexes of each object, turning them, and trying to line everything back up by eye - when all I had to do was click a button.

So ... talking about doing it wrong, there's more.

I've been going about making my game wrong. Or avoiding making it would be more accurate. This is partly - or even mostly -  due to a helpful blogpost by a developer that talks about Steam. Why is it helpful but made me lose focus on actual development? Because of wishlists on Steam and drumming up future sales. He pointed out that his previous and very successful games had a Steam store presence 1.5 years prior to launch which meant that he had managed to accrue 33,000 wishlists. So when he was finally ready to ship, 33K people got an email telling them it was available. Even steam suggests getting at least 50K before release.

I have half my levels completed, all my monsters for each level save the bosses, and complete working game code that makes it all playable even if some pieces are missing. But I don't have anything public facing. I have no cool box art or even working main menu screen. I could have done this previously but I was too fixated on making the actual game work that I never even took into account that I needed to get potential player support early on and that means having a working Steam store page with nice artwork that brings people in, adds backstory and concept. I don't even have a title for my game other than the temporary working one. In fact my previous game; Airship Dragoon had no final title until the month it went on sale.

Here is a link to the full article. Here's another one on wishlists.

So ... doing it wrong ...

The Clip Show Episode Except It Isn't

It's been a hectic month. So hectic I'm going to have to make a filler blog post. But first some actually useful progress ...

I was helping someone out who was having issues getting the exported collision meshes to work whilst using the imported FBX format from Blender 2.8, to the open source Torque3D MIT Version 4; Preview Build 5.

Binaries of which are downloadable here:
http://ghc-games.com/SC/Preview4_0_bld5_win.zip http://ghc-games.com/SC/Preview4_0_bld5_lin.zip http://ghc-games.com/SC/Preview4_0_bld5_mac.zip

Or at least I thought that was what the issue was. It might have been something else. Either way, I successfully resolved this issue, real or imaginery, so I may as well take the time to share it.

 The secret is to use the same node system as you would for DAE (COLLADA).

It appears that when importing FBX format naming is more required than when just using DAE. Anyhow I had a play around with exporting FBX from Blender and here's what I found.

1: It really likes the DAE hierarchy if you want working collision meshes when exporting FBX.



2: If you import as SingleSize your object meshes require trailingNumber LOD else the collision meshes will not work. Weird right?

3: It will import working collision meshes when using DetectDTS or TrailingNumber (even when you have no trailingNumber) without trailing numbers for LOD. Did I mention trailing number in that sentence at all?

4: FBX is HUUUUUUUUUUUUUUUUUGE. Importing at scale 0.01 is the same as importing a DAE at 1.0.

So, on to the filler bit.

Headpats ...

In other news I ventured into the 21st Century and got the cheapest fibre optic available, finally allowing me to upload files 9 times faster than attaching them to the leg of a dead pigeon and sticking them in the mail box.

Here is a gameplay video which was 2 gigabytes that uploaded in 30 minutes. You still have to manually select 1080HD on YouTube or else it'll play the 480 pixel version, even if you got full screen and it lies to you about which quality version it's playing ...

So, that was the month the wasn't ...

Big Boss Battles

The Big Lad

So after the joy/struggle/despair/awkwardness* (*delete as applicable) of learning how the new Blender3D works and then creating an actual character to replace the player's Small Yellow Placeholder Cube, it was time to go after the remaining Big Red Placeholder Cube which had been standing in for the level boss character.

As ever things went horribly wrong for seemingly know apparent reason other than the entire universe having it in for me.

I had split up animations from the actual mesh - that way it keeps loading times down whilst changing things on the fly during development, and means that multiple models with the same armature skeleton can share animation data and thus reduce file sizes - when suddenly everything stopped playing nicely and went back to mutating into a horrendous mess of twisted limbs and multiplied rotations.

The reason this time turned out to be because I was exporting the mesh model in the first frame of the root animation - which is what you would want for blended animations - but I should have exported it out in a copy of the rest pose ... regardless of the fact that this had previously been fine with another model ... which meant that the solution was far down my list of things to test to see if it worked ...

Exporting COLLADA/DAE format from Blender to Torque has become even more specific in Blender 2.8, mostly due to there being a huge number more options available and thus a huge number more things to possibly go wrong ...

I before E except after C ... Also Hierarchy which breaks the rule, and hierarchy is very important ...

I ended writing a little explanation of Blender3D to Torque3D Collada exporting and the importance of hierarchy over on the forums.

Eventually I managed to get "The Big Lad" - as he is currently known unless I can think up a name later - into the game and working. Here's a quick look at him in action. I say "quick" because he squashed me in not too much time. Select quality 720x60fps manually or youtube's autoHD says it's selected it but it's telling fibs.

I fearsome foe! Especially as I end up as pate ...

Apart from that, I modified my laser-beam-wielding veclociraptors by adding helmets and power-packs to them so that they now look different from the other dinosaurs which just bite.

Sharks Dinosaurs, with frikken laser beams on their heads!

I also spent some time modelling industrial scenery, lots of pipes and stuff for a forthcoming level. This will be the fifth level in the game, out of a total of ten - though they will loop for long play. No actual pictures of that yet as it is still all in bits.

So, that was the month that was. It got really warm and now Autumn has truly arrived and I've put socks on. Lot's of other things happened game development-wise but I can't really remember them. Next up is the Overgrown Industrial Level which has been reclaimed by nature, and more boss monsters.

Placeholder Yellow Cube Has Been Retired

Don't mind me, I'm just retiring a yellow cube.

If you want to be pedantic, it's actually a rectangle.

And here's it's replacement going all John Woo/Chow Yung Fat/Revy with dual wielding pistols.

Ooooh looks gamey!

But first to Blender3D 2.8. It's the newest version of Blender3D with a whole host of changes - like inverting mouse buttons for selection ... which is gonna take some getting used to after 15 years of it being the other way ...

However it did have a new addon I wanted to test out, an addon for generating LODs (level of detail) for my high poly mesh by creating retopology of the mesh automatically. What I wasn't expecting was for the new Blender 2.8 to hunt down and retire my older version of 2.78c which was in a totally different directory location.

So it turned out to be time to learn the new Blender whether I really wanted to or not ...

This was going to happen anyway due to the new Blender being all PBR and stuff ... it's just that I wanted to worry about that later, but now I had to worry about that now. The first problem was ... well ... PBR, and the fact that no materials or textures would render without creating some sort of spaghetti of virtual nodes - and this was just to get textures to show up on the model. And of course PBR textures won't render without both roughness and metal values.

Spagbol junction in the left pane, just to make textures visible in the right ...

After a bit of work, et voila!

 And how it looks in the latest, PBR friendly version of the open source game engine.

Catchick hanging with the TripleA Thicc Squad in PBR land

Which is when I found out that transparency was broken and filed an issue report on the code repository, github ... by which I mean I whined in the discord server to the power's that be, because github is my mortal enemy for reasons of never having been able to successfully create a working pull request that didn't get rejected ...

So, back to that auto-retopology LOD generator I wanted to test. It's called Game Asset Generator and is looks like a pretty nice piece of kit. I had a model with around 27K triangles and beefed it up to a completely ridiculous 440K using some catfish-bob subsurfing (or whatever the hell it's called).

From "USE ALL THE TRIANGLES!!!!" to "use a more reasonable number of triangles in one simple click ...

The real down side to it was that it completely mangles the original UV maps into one horrible mess, and these were something which I wanted to try and preserve.

Whisky Tango Foxtrot on the left, as opposed to the original on the right

Now this is why people go about the arduous and rather boring process of manually retopology on high resolution meshes. But I want some sort of quick fix and my cake and eating that cake and still having cake, so I set about a cunning plan and broke the model up into separate sections based on the UV island layout, before automagically creating a lower topology. I then checked that the seams were still intact and adding new ones if they were not and recreated an approximation of the UV map, stitched all the parts together again and finally baked the textures from the original high poly mesh. Whew ...

And that all worked quite nicely. It wasn't "quite" as automated as I would have liked, and wasn't "quite" as good at retaining mesh loops as I had hoped for, but it did work. Having said that perhaps taking the extra time to manually retopology the high poly mesh into a lower poly mesh may be worth it ...

Next up was to rig the whole model with an armature for animation, and export as we would normally via the COLLADA exporter. Except the new Blender3D 2.8's COLLADA exporter has a whole range of new and completely undocumented options. So, in usual fashion I just went and dived in testing what worked and what didn't.

 Nope

Nah

Eventually I got a rigged armature exported. It turned out it didn't like being parented to the actual skeleton directly and could pick it all up via the armature modifier. Now for animations!

Dancing on ice? Close but no banana as far as a nice clean export goes ...

To successfully export an animated model required making disabling exporting of all actions and relying on keyframes only, whilst keeping bind info and sorting objects by name to prevent a duplicate armature from arriving for no particular reason ...

I wrote a more detailed thread up on the forums here.

And eventually I got a working, and fully animated character out of Blender3D 2.8 and into the game.

Those Taoist charms say no and they mean no

Whilst trying to get the more hardcoded movement-to-attack animations working I came across numerous issues (which is just kinda part of life as an indie dev to be honest, especially when you're a One Man Army). The first was an unexpected naming convention for attack/recoil animation which took a day and a half to solve. This would have been a lot quicker if I had just gone and looked through the source code instead of thinking that something else was wrong.

The second was that the weapon finite state machine (FSM) kept overriding my second fire state with the first. The character has two handguns and I wanted them to fire each in turn for the standard light attack, and then both together in a heavy attack, except only the first light attack would work and occasionally half the animation for the second would attempt to play. It turned out that because I was using a hardcoded "stateFire=true;" setting, on loading it would store the animation, audio and particles for the fire state into memory and reuse it in any other state that had stateFire. However, stateFire does not actually launch any attacks or projectiles as that is all done via each states script, so by disabling stateFire on the second light attack I could prevent it from overriding the state and have the left handgun shoot.

And here is the game in all of it's post-yellow cube glory. This is the old DirectX9 version and not the new PBR one, recorded at a slightly janky 30fps because my upstream internet is bad.

So, that was the month that was ... I learned the new Blender and PBR by accident, pummelled exporting and animation until it worked, and finally replaced placeholder cube with an actual working character. And tomorrow is the first day of meteorological autumn.

Goodbye Green Grid, Hello Darkness My Old Friend

I took a break from worrying about the minutiae of cat ears to do some level building. It's been a while - in fact around 4 years - of having a basic and rather blank grid for level 1. This has now been replaced with an actual level 1, containing ... you know, level 1 stuff.

 Old And Busted: Not Exactly Grey Box as Green Grid Box

New Hotness: Actual Art Assets 

This required actual models and textures and stuff, so I opened the long dormant folder of 3D models and 2D texture images which I have collected over the last nth years. One thing which I did get years and years ago was Forester Pro - a procedural tree and plant creator. Not only can you create a ton of different types of trees but create variations with a single click of a button. It automatically adds vertex paint which is then used in-game to add the shader effect for the wind. You can also add or remove elements like branches and leaves, though I did find that the save file dies horribly if you go too much off template. Never mind, just export out as DAE format that the Torque3D MIT Open Source Game Engine that I use, and of course can also be imported into Blender3D for making changes to the model more easily. Here I used the normal editor so that the foliage planes face upwards to avoid bad shadowing artifacts. I added an empty high above and had the normals aim for it using Directional mode, that way each plane gets a unified but still individual shadowing.

Wind Effect

Also good for grass 

 I have 10 levels planned, and the enemies all modelled and animated, though not the bosses. So I figured that I would need a few trees. One thing that I found was that the standard black alpha on textures didn't look very good. There is a free demasking tool which writes in colour information but instead I tried using different colours of the image to see what effect they had.

It transpires that alpha colour is important ... even though you can't see it ...

I also had a plant pack for Forester Pro and after exporting a few plants I amalgamated the textures into an Atlas texture to save on drawcalls, and created multiple plant objects to save on instancing.

Not seen here, mixed plant meshes

My next issue was lag. The first reason was some terrible maths which I had written and my code was attempting to spawn 1700 enemies all at once when it really only want 7. Yeah, that sort of thing is going to bring the computer core down to a shuddering halt. Next up was collision. I had set the size of terrain squares down to 0.25m which created a rather high 1568 collision calls per player, Ai and dynamic object. I played around with different settings and finally settled on a square size of 1, which seemed to give good definition visually to the terrain and only cost 128 collision calls which was a rather saner amount. SquareSize 2 would have only cost 32 calls but I preferred the look of 1 square per metre.

When you attempt to use ALL THE COLLISION CALLS

One thing which I had not previously thought about was the lighting and shadowing. Normally the camera looks through the environment and the shadows fade and lose clarity the further away they get. However using a top-down/isometric(ish) view meant that I could see shadow splits drawing across each other. A lot of fiddling with shadow settings later I finally got something that looked decent and transitions between shadow map splits were not really noticeable.

The joys of trial and error shadowMap creation with the helpful PSSM Cascade Visualizor debugging tool

As my first level is a graveyard featuring a very lazy mouse who is supposed to be the caretaker but rather neglects her duties in keeping the place tidy for her customers - level 1 enemy, the dead (DEEPEST LORE!) I opened up the old graveyard art pack which I bought years ago. Some of the model seem to have been rather hurriedly put together with a lot of opening faces - which is the sort of thing my OCD really doesn't like, so I spent a fair bit of time fixing them and amalgamating the textures into single Atlas textures for those big savings on drawcalls.

Blender3D's decimation tool got into the spirit of spookiness

I had noticed an issue with the terrain becoming washed out when the level reloaded. After much hunting around for bugs in the code I discovered that I had accidentally changed the file format for saving the terrain main texture to a low definition jpeg of only 90Kb. The default is a high quality DDS image so I must have been playing around, not noticed the difference in texture because it only updates on reloading the level, and then just forgot about it. Whoopsie!

Adding grass meshes that sway in the wind make the flat texture look a lot better

So, next up is furnishing and finalising the rest of the level and making it playable. The either on to the other levels or back to modelling catgirls.

Head pats!

Terrible Tentacle Terror!

So, I managed about 3/4 of Dry January before falling off the wagon. Not a record by any stretch of the imagination but there was a solid few weeks without any booze in there.

Back to gamedev ...

In the last month of the last year I left on the nail-biting cliffhanger of ... modeling yet another enemy. So yeah, 3 years in, and the player is still a placeholder cube ... anyhow, back to that enemy character.

 Tentacles, in gloriously awkward to animate 3D

Tentacles. The scourge of cartoon schoolgirls everywhere (okay, only in Japan - literally nowhere else in the world, it's just Japan).

Cthulhu moves to Japan ...

It turn out that one thing a big bundle of psuedopods has trouble with is avoiding clipping into and colliding with each other when you use the minimum amount of keyframes in an animation. Tentacles turned out to be not that difficult to animate, pose into a nice, organic tentacle shape one way, then n frames later and another keyframe the other way. Let Blender 3D do the "tweening" (which is an animation term apparently for automated process of animating between keyframes) and then tweak the middle frame if it looks funny. Unfortunately there was a lot more frames which ended up needing "tweaking" to stop the writhing mass of extremities becoming a jumble of clashing appendages.

Did I ever mention that everything in game dev takes way longer than it would first appear to? Oh, every blog post? Yep, right ...

Placeholder cube getting snug cuddles from tentacles

Anyhoo, eventually said tentacles where animated, calibrated, datablocked into actually enemy AI and unleashed into a game level for tweaking - and boy did they need some tweaking. Player ended up squished rather quickly, then it was a case of tweaking data from too easy to too hard until it met in the middle somewhere.

Big tentacle, little tentacle

During testing I noticed an issue. Every so often a tentacle would fall from the sky past the camera to where it was supposed to be in the XYZ of the game world. I've seen this happen with other models  before and each time I fixed it and then forgotten what the cause was. The last time I fixed it I remember thinking that next time I will know the cause adn solution ... except for I have forgotten it again ...

I went back and checked through my written notes on exporting animations and then through all my "how to" forum posts ... and it looks like I didn't record the reasons anywhere. It's got something to do with animation or COLLADA format export - but that's as good as my memory is helping right now ... so a bit more testing and tweaking of the model and exporter is required,

Tentacle surprise!

I also made a whole load of really cool effects, like spawning and fleeing and attacking - but in actual playtesting there was so much going on that a lot of it is easily missed and the constant rattle of tentacles bursting through the ground gets rather annoying, so I plan on simplifying the audio and also some of the particle effects.

So, that's pretty much January covered. There was some other maths related stuff concerning getting player evasion to trigger the way the player is moving when it's different from the way they are facing - which ended up being more difficult than expected because I forgot to take randomized camera rotation into account on spawning and also forgot about normalizing vectors in 3D space - but it was mostly tentacle animating.

To be continued ... possibly indefinately at this rate ... naw I'm just kidding, I'll ship before Half Life 3.

Glowing Golems And The Curious Tale Of The Non-Invertable Bind Shape Matrix

Whilst agonizing over golem design followed by golem re-design, I had a mangled LOD mesh which resulted in a crash. Making a slight change to the mesh and re-exporting fixed the crash, but I also noticed a curious error warning in the console (which probably wasn't related).

bind_shape_matrix in mesh# controller is not invertible (may cause problems with skinning)

 Wut?

 A little bit of digging around the codebase led to these most useful code comments.

   // Some exporters (AutoDesk) generate invalid bind_shape matrices. Warn if
   // the bind_shape_matrix is not invertible.

Wut? 

 
Okay that wasn't very helpful. What would the skinning problem be (I had not noticed any) and how do I fix it for export from Blender3D? Alas, the interwebz were not particularly useful in  determining how to do this, so I started a process of elimination, based on the logic available. Deducing that many of my models were rotated and vastly resized soley using the armature, I eventually found that uncoupling the mesh from the skeleton, and manually editing the mesh to resizing, rotate and align it to the bones cleaned up this error message.

The downside turned out to be how many instances of this error there was. I have so far created a variety of characters which weighed in at ~110 seperate meshes, and 85 of them needed resizing and manually aligning - and as it was the meshes which required directly editing, I couldn't just snap one object to another - which is actually part of what caused the problem in the first place. At least it was an error caught before I got to the end of character creation, but it would have been much less hassle to have found this out at the beginning, possibly with some huge flashing warning than a few red lines in hidden amongst the jumble of text in the console soley during first importation.

Golems, golems a plenty

So this brings us to golems, and the constant redesign of the little blighters. Who would have thought that creating a vaguely humanoid shape out of clay/rock/earth would have been so complicated. The problem was, I didn't really like any of the designs which I came up with, thus creating a whole array of slightly different clay men ensued. Eventaully I went back to my first design and added glowing cracks. Initially I had these cracks animated, first with a pulsing effect which didn't seem to be very obvious, so I changed that to an animated one. This looked quite cool when they were not moving but became rather busy and not particularly obvious with them running about, so in the end I decided to keep the glowing cracks static.

I made a rather interesting spawn animation from which the golems are created from animated rocks gathering together and then fall away to reveal the monster within. It looked quite fancy having it happen over time but this is an action game so the whole thing had to be compressed into one second. Likewise when a golem is slain it collapses into it's constituent parts of rock.

Golems Spawing, pre-glowing cracks

In game, the golem is a small and heavy enemy, best suited to swarming the player. As they are made from earth/rock/stone/etc they are very heavy and cannot be pushed out of the way (at least not by the physically weak gun wielding, range-based default player character). Golem has death in Hebrew written on it's face - not that you can see this at the scale and angle, but hey it does.

I also changed the overall colour scheme of the game, getting rid of the greenish colour and replacing all of the HUD elements and user interface with an old blue colour I liked.

Green out, blue in

And here's  the golem level in action.

So far I have around 70% of enemies for levels done - not including level bosses, and of course there are a number of player characters to yet create and replace that yellow cube. I am leaving those until last as they are all a variation on a theme and will share the same skeletal structure for animation.

Next up is organizing animations and data for attacks for my old "Dimensional Knights".

Soon ...