cloth solver not as robut as maya's nucleus?

[kelvincai]

I am pretty new to houdini. Please bear with me if I missed something obvious.

I am try to re-do the car crash scene as in the side effect's cloth tutorials.

I try to build it from scratch using one of the evermotion's car model and use ray/isooffset to build a proxy (13k polys) for sim.

I couldn't get houdini's cloth solver to converge. But, when I run the same proxy in maya's nucleus engine, it runs pretty okay.

I think I am getting something wrong here, as houdini's cloth solver SHOULD be more or as robust as nucleus. Experts, please help!

[kelvincai]

I am pretty new to houdini. Please bear with me if I missed something obvious.

I am try to re-do the car crash scene as in the side effect's cloth tutorials.

I try to build it from scratch using one of the evermotion's car model and use ray/isooffset to build a proxy (13k polys) for sim.

I couldn't get houdini's cloth solver to converge. But, when I run the same proxy in maya's nucleus engine, it runs pretty okay.

I think I am getting something wrong here, as houdini's cloth solver SHOULD be more or as robust as nucleus. Experts, please help!

ecar_error.fbx.hip.zip

[brianburke]

I think I am getting something wrong here, as houdini's cloth solver SHOULD be more or as robust as nucleus.

Unfortunately I don't think this is quite the case yet. I think the H10 cloth stuff has some catching up to do performance-wise.

H11!

[michiel]

I suspect that you're not actually simulating the proxy.

If you change "select input" from 1 to 0 in "switch1" inside "ecar_lowpoly_fbx",

then you get simpler geometry, on which the simulation runs just fine in the

latest build of Houdini 10.0.

Is this the geometry that you were intending to simulate?

[kelvincai]

I suspect that you're not actually simulating the proxy.

If you change "select input" from 1 to 0 in "switch1" inside "ecar_lowpoly_fbx",

then you get simpler geometry, on which the simulation runs just fine in the

latest build of Houdini 10.0.

Is this the geometry that you were intending to simulate?

I make the switch1=0 proxy(~1k poly) to have at least something out from the sim and test.

The switch1=1 proxy (~13k poly) cannot be converged in my ability.

The original/to-be-render geo has over 100k poly.

[kelvincai]

Unfortunately I don't think this is quite the case yet. I think the H10 cloth stuff has some catching up to do performance-wise.

H11!

nucleus (as many realtime engines) is position-based and explicit, while houdini is force-based and implicit. houdini should be more stable (able to converge high stiffness) and accurate (follow physical laws), but slower in each timestep. However, to achieve the same level of realism, an implicit force-based solver should take shorter or equal amount of time than an explicit position-based solver, if the explicit can still converge.

it is a bit like the biased vs unbiased renders in the GI realm. maxwell (unbiased based on MLT) is more accurate but slow to get noise-free solution, vs mr, vray, fr, are faster but less accurate.

how could I debug how/what went wrong of the integrator in the console/prompt?

Edited April 5, 2010 by kelvincai

[kelvincai]

I can't get the sim to converge with the proxy over 5k poly and the high stiffness to make it metal like.