Softbody collision problem - scale?

[brainjuice]

We're working on a serious simulation in the medical area. It involves interaction between soft tissue and rigid immovable bodies... e.g a device pressing into living tissue.

The problems we are having is that collision between the softbodies and the rigid collision shapes (bullet primitives like spheres and boxes) are very inaccurate.

One thing we discussed is that the scale of our world is small... about 1 unit across which means our collision shapes are of the order 0.01 to 0.1 units in dimension. The coder we've had working on this says this is comparable to the default collision tolerences in bullet which would explain the lack of accuracy. We've not been able to establish if we can lower that tolerence relative to our world's size, or if there are numerical limits beyond which things will start to fall apart?

Thanks much.

See here for paid gig post: viewtopic.php?t=3990

[Erwin Coumans]

The soft body collision detection is not very accurate indeed. Have you tried to scale up the world by a factor of 10 or so?

By default, collision detection only happens at the nodes/vertices. Another option is to use collision clusters, have you tried that?
For better accuracy, it might be better to implement per-triangle collision detection:using a single cluster for each triangle.

Thanks,
Erwin