Real-Time Physics Simulation Forum

Going over the PGS paper by Catto, I noticed the main reason for using weight dependent (as opposed to contact force) friction was used to simplify the solver for friction constraints by removing the interdependence between contact and friction. In the case of an iterative solution solver that can probably save quite a bit of time, and general use prioritizes speed over accuracy. But for the case of a direct solver, is the interdependence an obstacle to finding a solution in reasonable time, or is solving still complicated by that interdependence?

Most physics engine I know, including Bullet, use the normal force for friction, approximated as a friction pyramid. When using an iterative solver such as PGS you can re-compute the maximum friction forces at each PGS iteration.

It is also possible to formulate the friction as part of the MLCP, see function (54) in our Eurographics state of the art paper (https://code.google.com/p/bullet/downlo ... e=STAR.pdf) . The resulting LCP cannot be solved using PGS, so you need an MLCP solver such as Lemke or PATH. You can see an example implementation of this friction model here: https://github.com/erwincoumans/daVincy ... d.cpp#L330