Head joints in AF file outdated?

[Ishtvan]

I just noticed that when ragdolled, our AI's heads aren't following the ragdoll physics, instead they just stick straight out stiffly from the torso all the time. I tested Builder guards specifically, believe it also happens to pro guards.

 

I'm pretty sure it worked correctly at one point. The heads should flop around when you hit them. It's not an AF physics problem, I can see the head CM flopping around, but it's not modifying the visible head model. This probably means the AF file has to be updated for new joint names in the new skeleton. The old skeleton AF and new skeleton AF both have this same setup for the head:

 

body "head" 
{
joint "Head"
mod orientation
model box( ( -4, -3, -6 ), ( 4, 3, 3.5 ) )
origin ( -2, 0, 75 )
density 0.4
selfCollision 1
containedJoints "*neck"
}

 

Are there still "Head" and "neck" joints on the new skeleton? More importantly, what joint is the head actually def_attached to? Perhaps this joint is the one that's not being modified by the AF physics? It should be whatever's in the head_jointname spawnarg, right? Is it not "Head" for all our AI?

[Springheel]

The head joint has changed. On the new skeleton, heads are attached to Spine2.

[Ishtvan]

Okay, I changed it to Spine2 in the AF file and it looks like it is working again. The ragdoll looks more believable now.

[Ishtvan]

Argh! Now the ragdolls of guards with helmets are messed up. The helmet seems to be solid to the AF, so it stops the head from falling down as it should. We've probably always had this problem, but fixing the AF head physics made it more noticeable.

 

Can we try setting the helmet nonsolid on death or to CONTENTS_RENDERMODEL or something?

[Springheel]

Ok, I added:

 

"drop_set_nonsolid" "1" // will become nonsolid when owner goes ragdoll

 

to the helmet entities.

[Ishtvan]

Hmm, I have another solution I just tried. I got rid of CONTENTS_CORPSE in their clipmodel contens and just set them to CONTENTS_RENDERMODEL.

 

I teste this and it worked. They shouldn't have to be CONTENTS_CORPSE. Only reason is for the melee initial collision. If they're not CONTENTS_CORPSE, the initial collision will hit the head model, but then the next trace should detect the helmet. This is more iffy in principle, because you can hit the CM at a weird angle such that the next trace sometimes doesn't hit the surface.

 

I'll try your solution of setting nonsolid. It would mean that arrows/blows will react as if there's no helmet on a KO'd AI, but maybe that's okay. Ideally we would have a drop_set_contents_rendermodel as well, not sure if we have that already.

 

The real reason that CONTENTS_CORPSE is a problem is: It adds the helmet to the AF, which is fine, but it sets AF self collisions to 1 by default on anything added to the AF. This works for swords and arrows, but on the helmet, this is bad, because the head is constantly colliding with the helmet and stopping it from moving. So ideally, what we would want is a spawnarg to control whether self collisions is 1 or 0 when we add the item to the AF. That's a more complicated fix though, that I don't think we should try right now.

[Ishtvan]

Actually, drop_set_nonsolid doesn't look like it works in this case, judging from the code:

 

	if (ent->spawnArgs.GetBool( "drop_set_nonsolid" ))
	{
		int curContents = ent->GetPhysics()->GetContents();

		ent->GetPhysics()->SetContents(curContents & ~(CONTENTS_SOLID|CONTENTS_SOLID));
	}

 

It clears the solid flag, but it leaves the CONTENTS_CORPSE flag in place. I might add a new option like drop_set_corpse_nonsolid that also clears the corpse flag. Or just drop_set_contents_rendermodel. That should be good enough for now.

[Ishtvan]

Nope, even with that fixed, it's already too late by the time drop_set_* stuff is called. The AF representation has already been added, and that's what is causing the problem.

[Ishtvan]

I take that back, it's not getting added to the AF because its CM is not a tracemodel, so that's not the problem. Something else weird is going on. The only thing that's worked ragdoll-wise so far has been setting it to only CONTENTS_RENDERMODEL from the beginning.

[Ishtvan]

Oh, Duh! This isn't working because the helmet is not attached to the AI, it's attached to the head that's attached to the AI. Argh.

[Ishtvan]

Okay, added code to also set attachments of attachments nonsolid, but now I realize that the head isn't even in this list of attachments that's effected by drop_set_nonsolid, because it's handled differently than other attachments. Double-ARGH!

[Ishtvan]

Okay, attachments to the head are now treated the same as attachments to the body, so the same spawnargs apply. With the exception of melee weapon attachment-related spawnargs. I was too lazy to add CMeleeWeapon dependencies to idAFAttachment, but I figure we won't have any AI with a dagger stuck to their head, or any head butting with lethal pointy helmets any time soon.

 

(In the code, I added a DropOnRagdoll function to idAFAttachment class, and had the AI call it on their head when their own DropOnRagdoll was called).

 

Ragdolls with helmets now WORK. This was another reminder of how a 5 minute fix (make the AF file refer to the correct joints) turns into a 2 hour fix. Can of worms opened, and closed (mostly closed, attachments to attachments that aren't heads are still an issue for the future).

[Springheel]

Ragdolls with helmets now WORK. This was another reminder of how a 5 minute fix (make the AF file refer to the correct joints) turns into a 2 hour fix.

 

 

And to think, I almost posted, "well that was easy" after your second post.

[Ishtvan]

 

And to think, I almost posted, "well that was easy" after your second post.

Yes, those are famous last words, I've found.