Originally Posted by ThE_JacO
Sorry, but what system do you know of that uses non linear springs?
Beside the two facts that A) deflection is enormously expensive to compute in a large set of connected nodes B) a good spring is as close to linear as possible, a rubbery look doesn't come from the spring model, but usually from the constraints within the network.
It's also hard to impossible to simulate thickness cheaply in classic SBDS, since you would need a very carefully placed topology for the hull, and an additional set of springs connecting the two sides that would transmit force instead of constrain the network.
Again, what system have you seen that elegantly does that with arbitrary topology? I know/have seen a few that can do something with a tetra pass, and some CAE models, but in DCC?
I think you are blaming the model (spring networks) when it's actually perfectly fine, when instead you should look at whether you have or not all the options to configure it properly (angular and length constraints, proper dampening control and so on).
ThE_JacO, and everyone else:
No disrespect man, but I'm a physicist turned artist, and I think we are probably more kindred spirits than you realize. I coded with non linear springs way back in 92 (on a bridge simulation project at one of our national supercomputer centers) so I know it's not that difficult to do and worth the effort. Here's an old (but great) Siggraph 88 paper that outlines a nice softbody system: http://graphics.stanford.edu/course...erzopoulos2.pdf
They had spring & damping forces (like we do), but also plasticity & fracture going on as well, and the results were beautiful and well worth a look. I'm a little touchy on the subject because this work is over 30 years old and still kicks our soft body deformation system's backside in many ways IMO.
Yes the rubbery look does mostly come from the spring model...
The root problem is that springs are a great first step, but only first step because real materials simply don't act like perfect springs. FYI, the "perfect springs" many programmer types learned about in school in basic physics classes are really just theoretical fiction used to illustrate a concept. Real springs are remarkably linear, but only over a certain range. Everything in fact resists deformation like perfect springs, but only over a certain and usally quite small range. If you just want to simulate small deformations, then linear is great, but if you need a fair amount of deformation on supposedly non rubber objects, linear springs don't cut it.
Have you ever seen cloth jiggle or stretch like a spring?
Yes LW damping works great to reduce jiggle (by a process called critical damping in the real world), but doesn't touch the problem of all of our soft bodies deforming like cartoon rubber.
Forces are simply not great enough to resist large deformations accurately. Using damping in a heavy handed manner to try to reduce the amount of deformation produces problems of it's own.
Also, linear spring based compression has compression problems...dense meshes can easily get "turned inside out" in areas for example...for example, if a corner of a soft body cube gets moved too far in the direction of the cubes center for example, it can get stuck in that dent-like position.
Yes, auto grid internal simulation geo is maybe too much to ask for, and maybe unrealistic from a certain pov, but this is the solution for how regular artists can get get non cartoon rubber deformations... Gridding has been around a while in FEA/Engineering type application, and wouldn't be all that difficult to at least come out with a very basic implementation.
Adding non linear springs is low hanging fruit however and should be implemented now.
Heck we should have had it a long time ago. Doesn't really matter that other vendors don't have it yet, in fact it's an opportunity. I might be crazy, but not about this anyway.