This paper starts by describing how collisions are a hard problem
and first describes what inpulse-based simulation methods are
and how they can be computed efficiently. Showing the comparison
of the stable pose distribution of a wierd looking piece from
real experiments and simulation, it shows how impulse-methods
can portray realisitc movements fairly well. However, it shows
how constant calculation that is not required for steady state
objects can increase computational load and constant constraint
situations like hinge joints that can't be effectively solved
using impulse-methods and hence require constriant-based-methods.
The paper suggest hybird of impulse-methods and constraint-methods
can be used to effectively simulate physical phynominas.
However, it doesn't seem to be able to give a clear picture on
when and how switching between constraints and impulses should
take place to be effective and leaves the question open.