[ODE] possible bug?

[Russ Smith]

> I have found that I can cause an LCP error with two boxes sitting
> adjacent to each other:

the problem with your example is that the vertices of the two boxes
line up on the ground. co-incident contact points cause a large loss
of precision in the factorizer/LCP solver, which is what is causing
the error message. if one box was moved/rotated slightly you would
not have this problem. BTW, detecting and correcting coincident
contact points is a lot of trouble ... so you just have to be careful.

> Note that if the second box is placed at (1, 0, 0.5) instead of (-1,
> 0, 0.5) the error does not occur.

this is a coincidence of the numerics. in this case a contact between
the two boxes is not created (it *is* created in the former case).

russ.

--
Russell Smith
http://www.q12.org