15 - Self-collisions, solving the hardest problem in animation

[@kunqian7069]

Look forward to a tutorial on cloth tearing and remesh

[@Madlion]

The substep I think can be spatially variable, by maintaining a grid structure for average velocities in space, with this we can vary substep based on where there's most activity

[@seeyouinthenextlifeprobably]

It's much simpler than traditional methods, which is nice!

[@aydencook03]

Will you eventually make a video on a general purpose constraint solver? One that can solve any kind of constraint in their fundamental form instead of manually having to calculate gradients, etc .

[@mytino]

Great video! I've been using the friction model from your paper "Unified Particle Physics for Real-Time Applications". Do you still recommend that model? Is the "unconditionally stable friction" in this video a different kind of friction; more of a general damping? Could this perhaps even be used to do viscosity in fluids, or is that still better to keep velocity-based? (I see viscosity is velocity-based in your fluid solver challenge code.)

[@dnairlan]

Great tutorial Matty, thanks for sharing it! One question. Is it possible to plug your own integrator when using this approach or it is only PBD? Have you ever trying using e.g. an implicit integrator (e.g. Baraff implicit integrator) while using this approach?

[@TenMinutePhysics]

Thx :-) You could certainly use the hash grid with another integrator. The idea of replacing iterations by sub-steps probably does not work well with a standard implicit integrator. Also, the way I simply push particles to resolve collisions and the unconditionally stable friction I describe are XPBD specific.

[@dnairlan]

@@TenMinutePhysics I see, thanks. :)

[@okifunearl8511]

​@@TenMinutePhysics Great video! Could you please explain why the collision handling method and unconditionally stable friction method is **specific** to XPBD? Do you mean they only suit on small sub-step iteration, or there's other features that XPBD own but I missed. thx!

[@blastbuilder2430]

Great video! I'm wondering if you will do a video about the SDF method?

[@jayanth9661]

Hey... I really like all your videos. Can you please make a video which does simulation related fluid dynamics using particle fluids/SPH ?

[@TenMinutePhysics]

I will certainly do fluids at some point because I like fluids a lot :-)

[@АлександрЯкимов-к3о]

Hello, thank you for this great video! I'd just played with the demo. It is still easy to create some self intersections between cloth with a mouse. So probably not all self collisions can be avoided here.

[@matteocampo1732]

Great video as always. I was wondering, do you still consider the use of air meshes to keep the correct ordering of the layers?

[@TenMinutePhysics]

Yes, I plan to revisit them and will also make a video about them

[@anomyymi0108]

Thank you very much for these videos! I can't unsee your name in every other paper I stumble accross now. Do you have any resources to recommend on how to improve this sim further? I have some ideas but I'll see what people smarter than me have come up with first. Particularly on improving edge collisions (without throwing more particles at it) and making it simd/multithreadable?

[@alengm]

Thank you for these tutorials and for PBD in general. It's the first physics simulation method I was able to understand! Question. Do you have any recommendations on how to avoid sponginess in fluid simulations without making number of substeps proportional to length of the fluid volume? Grid based methods can ensure that the information about pressure propagates across all of the fluid but in PBD the information travels only through particles getting displaced.

[@Spedley_2142]

I was trying to use Beziers line drawing algorithm but as a time step rather than distance. Fast moving objects were stepped repeatedly and slow moving objects rarely updated. Everything was in finite steps. i.e. a bullet would be updated (in memory, not graphically) at 1000 times per seconds and a person updated 10 times per second, each moving a unit of 1 per iteration, then being put back down the list (based on Bezier's). I failed miserably but learned a lot.

[@foggyhazel]

Hi Matthias, I found your video from 8 years ago talking about air mesh for collision handling. Is that still applicable here? I am looking for a method to solve self intersecting cloth due to character animation (eg. arm pit and cubital area) which comes up a lot in production

[@UncoveredTruths]

ooh very nice, thank you so much!

[@TenMinutePhysics]

Glad you like it!

[@voxelltech]

Awesome content! I also did cloth self collision but using binary radix tree BVH. Is there any reason why you use spatial hashing over BVHs? Or maybe the pro and con between the 2 of them haha.

[@TenMinutePhysics]

Thank you. The hash is just very simple to explain but a BVH tree is certainly an alternative. Replace it in the demo and lets see what is faster :-)

[@voxelltech]

@@TenMinutePhysics I see.. Thanks again!

[@okifunearl8511]

@@voxelltech hi~ I‘m a little confused about how to build BVH over a single mesh like one piece of cloth? imo the smallest unit in BVH leaf is a whole entity mesh instead of some triangle?

[@dexensh]

Hi Matthias, is such collision handling technique robust enough for volumetric deformable bodies?

[@LowLifeGraphicsProgrammer]

Hi, I wonder is there any efficient, parallelizable way to solve self-collision of a 2d polyline?

[@TenMinutePhysics]

I haven't thought about this problem. What is your application?

[@ZioBlu]

I had the same question :) Currently working on a hair system.

[@DrTheRich]

Does "Local Optimization for Robust Signed Distance Field Collision" work together with this? if yes will you cover it? I'm designing a system for real-time ship rigging. where there is a lot cases of collisions of cloth with thin geometry like ropes and masts, which normally doesn't give good results in traditional systems. since there is a lot of clipping and glitching when the quality is lower.

[@AHSEN.]

I'm so sorry I had to break the perfect 256 likes by adding another one 🙈

[@shulehr]

If thats the hardest problem in animation then its probably the easiest field

[@atomictraveller]

it's the most lucrative, so surely any productive methods will be obfuscated to preserve value. there are a whole bunch of fat dudes at the lodge ready to sit on it, which is why there are like 11k views for a method developed a decade ago.

[@DasAntiNaziBroetchen]

If you think it is easy, it means you have no idea what you are talking about.