Home Creators Posts Import Register Favorites Logout

Content

Someone was curious about seeing colliders on characters. Well here they are, shown on 2 different morphs sets on the female character. I counted them because I was curious as well. The colliders auto-adjust based on morphs and skinned vertex positions so they are generally accurate. As seen in the picture I skimped on the shoulders and arms a bit to save processing time. The arms don't use auto colliders which is why they look same across the 2 characters shown here. I'll fix these at some point so they work with more morphing. If you see clip through on shoulders or arms, this is why. Ideally I would like about 8-10X more colliders per character to make it work even better, but the processing power is just not there yet and Unity's Physx implementation needs some further work to make this feasible. The upcoming male character does not yet have these auto colliders so it will also be subject to some clipping in the next release (1.1).

One primary goal I had with this project was skin-accurate collision, which is why I am posting this. It sort of shows how much further it needs to go before that goal is achieved.

Files

Comments

Wm (edited)

Comment edits

2026-09-14 05:00 Instead of so many capsules, why not use mesh colliders? You can slice-up her mesh into sections and then make extremely low-poly versions of them. I realize that's a fair bit of work, but 794 seems unnecessarily high. I'd think you could do it with ~10% of that if you did. You can always keep capsules for small things like fingers and toes where precision isn't necessary. (I've used 100s of mesh colliders in my projects and the performance hit isn't as bad as it was many years ago when physics were done on CPU.)
2017-08-04 23:21

MeshedVR

Mesh colliders are terribly expensive to update shapes at runtime, even if this was broken up in many smaller sections. And if the mesh collider is attached to a non-kinematic rigidbody they must be convex. I did experiment with this early on and it was a no-go. Updating a low-poly version of the character made the app run at about 8-9fps. It is possibly Unity has improved on this so it might be worth some more experimentation. If these were static objects, mesh colliders would not be a problem, but these are moving characters and the mesh is constantly changing. It is much more efficient to use capsule or sphere colliders attached to rigidbodies that move their anchors with the skinned vertex positions. I agree this is very brute force, but short of using a custom physics solution (particle physics would be a much better approach), I'm stuck with Physx.

MeshedVR

Oh - p.s. - 1/2 the colliders are for the soft zones where there is generally 1 collider per 1 or 2 vertices on the mesh. This is what drives lips, genital area, breasts, and butt. That is why the density looks higher there.

Noneof Yourbusiness

You are truly a mad genius. I had tried the same thing with the mesh collider I mean, but for me it worked almost fairly well. For a sort of low-med poly character I managed to update the mesh collider every FixedUpdate and still get around 30-40 FPS in VR with a slightly more complex scene than you have (terrain, shadows, grass, Water4 and some particle effects), but like you said, the problem is collisions. No matter what I tried Unity refused to allow collision/trigger events with that mesh collider :|

Noneof Yourbusiness

BTW, would LOVE to see the code snippet that places the colliders :P THAT'S the real porn for coders XD

MeshedVR

The code to place the colliders is pretty simple. Each of these auto-colliders is essentially an additional joint attached back to one of the regular bones (this allows direct push feedback). The code to update the position of these colliders simply updates the connected anchor to the position of the skin vertex each frame so that it is properly adjusted. Then for sizing, the colliders size is only adjusted when the character is morphed. You could adjust size every frame, but this is more costly to CPU runtime. Size of each collider is based on 3-4 vertices on the mesh. Again this isn't perfect. It would be much better if there could be a collider for nearly every vertex just like there is for soft zones now. It would be a lot more accurate and there could be variable softness all over the body. That is what I want to achieve long term, but it would require about 8000-10000 collliders per character. And Unity cannot currently handle that, but there is some promise. Take a look at this (job system is not yet implemented by Unity, but some of the transform updates are): <a href="https://www.youtube.com/watch?v=AXUvnk7Jws4" rel="nofollow noopener" target="_blank">https://www.youtube.com/watch?v=AXUvnk7Jws4</a>

Project 3 VR

Interesting approach. Do you have separate bone for each collider? Mesh colliders are expensive if they are for multi vertex models. But you could create your own simple shapes and not render them but only use them for the colliders. I think that was the idea of Wm. So you add collider and change the mesh reference to another simple one. But if each collider is a separate bone (separate transform), that is a different story...

MeshedVR

Yeah - separate joint(bone) per collider that ties back to main character bone for feedback. I have tried many approaches. I also thought that if I put multiple colliders on the same joint and update those colliders to adjust to skinned vertices it would be more efficient, but it turns out that a multi-collider joint is terribly expensive to update at runtime. It seems to trigger some expensive recalc in Physx that make it unfeasible at runtime. So it is actually cheaper in Physx to make many micro joints(bones) that attach back to the main character bone, and limit each micro joint to a single collider. The cheapest of colliders in Physx is Sphere. Capsule is not much worse than sphere as it is 2 spheres collision cals + 1 small additional math operation. Mesh collision, however, is a lot more complex to calculate collision on, even if the mesh is small. I did consider using thin box colliders and aligning with faces, but that would be for 1 face to 1 collider only. Making lower poly mesh and using that for colliders also quickly ends up in collisions not on the actual rendered surface. Instead I have found capsule colliders work great for bodies due to general curviness of the surfaces. And they are fairly cheap. So for instance, the thigh shown here, has 19 capsule colliders to approximate the surface. It isn't perfect, but it seems to be best tradeoff of speed vs. accuracy that I can manage for now. Eventually I really do want full vertex accurate collision everywhere. I'm scheming up ways to make that happen.

Project 3 VR

Thanks for sharing your experience Meshed VR. I will launch my Patreon and do the same in a couple of weeks. My experience as a developer is that the believe that there is one and only correct way to do something is the worst way to go. It is great that we try different ways and use different approaches, as each has benefits and drawbacks. I will keep following your way of doing things as it is a very interesting one!

MeshedVR

Some of the things I tried and gave up on were years ago back with unity 4.X. They have improved a lot, especially in Physx space, so some of the approaches I'm taking may be dated and there may be better ways of doing things know. I try to experiment as much as possible, but it is quite hard to keep up and my time is limited! I encourage trying other ways and see how it works out. You might hit a jackpot. Best of luck with your launch. I will keep my eyes out for it!

MeshedVR

Wm - was the mesh shape changing all the those vertices every frame (like a skinned mesh would do) or was this a 'static' mesh collider that just moved position? It is the changing of vertex positions in the mesh that are the problem with mesh colliders not the moving of them. For me to use mesh colliders in this project I would have to update the meshes every frame due to how body shape deforms, even if I broke this down into many smaller mesh colliders. If I have some time, I will try an experiment or two to see if things are any better. The math behind sphere/capsule collisions is very simple. The math behind mesh collider collisions is much more extensive. Also - with a body shape, convex mesh colliders probably won't work all that well in terms of accuracy in areas around the joint. I would love to be proven wrong. I would love to be able to see accurate mesh colliders for skinned characters that operate at high frame rate.

MeshedVR

p.s. I will be adding some static hair options. Also - I must have done something odd with the hair shader. It used to not slow down when getting close but I also noticed this the other day. I think it might have to do with the new shadow system I'm using. I'll look into it. I'm also looking into another hair solution that looks like it offers styling options and might be cheaper on GPU.

MeshedVR

Not trying to shoot down your suggestions, it is just that I have tried most of these things over the course of this project and this is what I found works best for performance and accuracy of collision. The recalc of collider positions is done on a separate thread and is already very optimized, which is one reason this works here. The actual adjustment of the anchor, however, happens on the main thread because that can't currently be threaded (Unity is working on this however). Colliders are already optimized to ignore other irrelevant colliders, but I want to support the character touching itself. I had a physics framerate option that I have since removed. It allowed 45fps, 80fps, and 90fps. 45fps causes too many glitches, and due to how physics not being threaded it would really only help performance of every other frame. I found it works best if physics runs at same framerate as rest of engine (90fps in this case). Unity has promised some updates (job system) that are probably going to greatly improve performance in this app. Unity 2017.3 is timeframe for initial system. They might not have the physics components tied in for initial release. They are allowing threaded Transform updates in that initial release though.

matthew b

where is the new update ?

matthew b

where can i download the full game ?

FracturedWhisper7

I'm not sure if this is something you've already considered, but I had some thoughts on ways to adjust the distribution of the colliders so that you could get the flexibility on the arms without having to add more than there already are; think about the parts of the body that doesn't have a lot of body fat, the digits, hands, feet, shoulders and back... and the "anal" area [not sure if that would have an impact on its current functional state]. All this is obviously assuming your soft body setup is based on regions of the body, and not a part of the physics of the entire model, I honestly don't know for sure.

MeshedVR

There is only single download for now (version 1.0 & patch 1.0.1). You enter patreon keys on main start screen to unlock the content and then just enter VR. Version 1.1 will be out this month. It is well underway with many improvements, tweaks, and bug fixes.

MeshedVR

The hands, feet, toes, arms, etc. are already "hard" colliders and use a minimal set of capsule colliders that will closely mimic the skin surface. There are a few areas that need a few more to be better to prevent clipping. I used less in more rigid areas since the skin does not bend there. I don't think I can cut back without losing significant surface accuracy. I really actually want to use more to make it better.

matthew b

where is the download ?

MeshedVR

The keys are posted in the "Posts" area.

Marty Burke

what are my patreon keys?

Velocity

I have a proposal to a new feature for the motion controller When you grab her hand with the grip button, you can make the girl close her hand into a fist the harder you press in the trigger., basically making her grab her ass or boobs . and still keep the collision on the fingers and the skin you're or rather that she is grabbing to softly/firmly grab it.

MeshedVR

If you do full grab of hands now, the triggers will actually flex the fingers a big. I'll be improving this going forward. I have plans for a touch feedback system and also some tweaks to the finger joints to make them behave better. I would like fingers to stick to things if pressed firmly. Currently they slide a round a bit. So I have both feature to add and some tweaks to make. Some of these will be in upcoming 1.1, but most will be in 1.2 or later.

PUTZ (edited)

Comment edits

2026-09-14 05:00 Is there still a place for Leap Motion after having Touch Controllers? Also if you could get all camera controlls mapped to the left or right controller in play mode would be great. Right now you have to use right one for turning and altitude. Unless im missing the way to link my view to one of the floating camera nodes? on same note can i tilt my camera down without having to tilt my head?
2017-08-19 03:27

MeshedVR

I do plan on looking into limited Leap Motion support. Virtual hands could be fun! Someone else asked for way to swap Touch controller function. I was going to all you to press thumbstick to be able to swap the function of the the thumbstick. In this way you could use either controller for full movement controls. Currently there is no way to link your view to one of the camera nodes, but I could consider that if I add 'tilt' support, which is you next question. There is no tilt support at the moment. I figured people would make scenes oriented the way they want, but I can see it might be nice to have this. Let me think about how to do that without causing some really odd navigation issues.

Adeptus Steve

Looking very interesting mate, I was wondering a few things and wanted to ask if youd be interested in a little chat from dev to dev. Let me know :)

MeshedVR

Absolutely! Love chatting with other devs. You project looks amazing! I have to find some time to check it out in more detail!

Adeptus Steve

Nice, cheers mate, maybe message me directly so we can share discord info or something?

Jerberjer

As this project goes on, will there be support/documentation for adding your own characters? I'm really interested in what you have here because of the skin-level physics on the whole body

MeshedVR

Currently character meshes and bone->skin rigging are based on DAZ3D genesis 2 and I will be adding support for genesis 3 in the future. I would like to be able to add support for custom morphs and texture sets, but they will have to be compatible with one of these 2 models for this to work. I can't support any other mesh topology as the mapping to soft body physics is a manual, time-consuming process. Same goes for the auto-collider system. Also - currently the soft physics is limited to several zones and is not whole body. The system is too expensive (CPU) to apply to all areas of the body right now.