Steven J. Barnett

For friction, I ended up applying simple Newtonian impulses for before starting the constraint solver. This saves some computation inside the PGS iterations (which are the biggest bottleneck), and not much is really lost since friction doesn’t need to be precise to maintain realism. I got this idea from reading the Box2D source code from Erin Catto.

Next step will be to use the cached normal constraint impulse values to scale the friction (the Coulomb model). Once that’s done, it should be far more difficult to pull out the lowest jenga block.