Re: [Algorithms] Collision on slow rigid body which cannot rotate

[Chris H. <ch...@d6...>]

>There's something about this approach that bothers me : constraints have
>nothing to do with the inertia tensor in the first place. Artificially
>linking one to the other sounds dubious. Tweaking the tensor might work to
>prevent rotations, but I think it's the wrong way.

This is not true, actually, when you really get down into the theory of 
generalized coordinates and constraints and whatnot.  In fact, constraints 
are exactly the things that change the inertia properites of a body; that's 
how they work.  If you pull on an object in a direction that it's 
constrained in, you feel infinite inertia in that direction.  The math is 
identical, and the theory is consistent.  In fact, you could make an 
argument that this is more theoretically sound than using something like 
Lagrange multiplier constraints, since LM is just a technique that uses a 
big expensive solve to figure out forces to exert to make the inertias come 
out infinite in some directions!  Why bother when the simpler way yeilds 
the same math and results?

So, don't worry, be happy!  You can claim theoretical correctness _and_ 
ease of implementation.  The only downside is it only works on the explicit 
DOF in your system (which is fine for single RBs, which is what we're 
talking about here), and they're only holonomic (but that's also fine for 
this application).

Chris