Squeezing a soft object simulation speed

[CarterBer1]

Hello, I am working on a project where we are using a rigid hand to squeeze a soft object. When we make simple contact with the soft object there is no issue, but when we try to squeeze it the simulation slows down to a very slow pace. We are using a simplified volumetric mesh for the soft object. I am wondering if there is any common considerations that cause the simulation to slow when the object is squeezed. Please let me know!

[bakpaul]

Hello !

Without seeing your scene file I can only make general statements on contact simulations.

If you are using Lagrangian based constraints for this then it seems coherent for me that the simulation takes time to be computed. The way we are solving constraint is based on a iterative algorithm, that suffers from bad conditioning of the system. When you squeeze something, you are imposing antagonists constraints, reducing drastically the conditioning of your system. The iterative algorithm will take a lot of time to converge.

I can think of three ways to quickly try to reduce this computation time :

1. System regularization: You could add some Rayleigh mass to regularize the system a bit. This will modify the dynamics of your simulated object but in the end, it might get better convergence.
2. Instead of using a LinearSolverConstraintCorrection which is the one that provides a system with the strongest coupling between constraints, you could use an UncoupledConstraintCorrection. But honestly, there are high chances that this will be unstable. You will gain in computation time but loose in stability
3. Try using the AugmentedLagrangianResponse instead. Here is an example : https://github.com/sofa-framework/sofa/blob/master/examples/Demos/fallingBeamAugmentedLagrangianCollision.scn. You can play with the epsilon to get a tradeoff between stability and computation speed

Of course, all of this holds only if your constraint problem is well constituted. Otherwise the main source of bad conditioning will come from the actual set of constraints if for instance some nodes are over-constrained (multiple constraints on the same dof that cannot be enforced together). You could also analyze your constraints to see if some points are overly constrained like this.