June 10, 2024

Field Notes from Power Plastics

Lessons learned while building a hybrid solver for mesoscale inelastic flows.

By Ziyin Qu

It has been a year since we began shaping the Power Plastics project into a cohesive solver. We started with an early constraint: artists wanted the expressiveness of material point methods, but our production partners refused to wait hours for previews. The compromise was a hybrid solver that advances particles in a Lagrangian sense while letting a lightweight Eulerian background enforce stability.

The turning point came when we stopped trying to perfectly match ground truth stress responses and instead optimized for energy consistency across resolution changes. With that perspective, we could aggressively decimate control grids without generating the wobbly artifacts that plagued the first prototypes.

Three small tricks proved indispensable:

Contact-aware sub-stepping so that fast plastic flows do not alias when colliding with thin geometry.
Implicit-thin shells on the Eulerian grid to keep the numerical viscosity in check even when animators layer dozens of forces.
A human-readable instrumentation layer—the shader team could literally paint where the simulation was losing volume, which saved weeks of guesswork.

If you’re hacking on related problems, I’d love to hear from you. I’m especially interested in coupling this approach with differentiable objectives so that we can co-design form and material at the same time.