Project detected nonadjacent cord-segment penetrations along the line
between their closest points, with Coulomb friction, in place on predicted
positions. One ordered pass does not certify separation of every pair.
A cord is a one-wide lattice, its particles in index order along it and
segment s joining particles s and s + 1. Each segment is a capsule of the
domain's thickness (its radius). For two segments, the closest points
p = a0 + s (a1 - a0) and q = b0 + t (b1 - b0) come from the clamped
segment-segment solution of Ericson ("Real-Time Collision Detection",
2005, section 5.1.9). The non-penetration constraint C = |p - q| - 2r
(r the radius) has gradients (1 - s) n and s n on a0 and a1 and
-(1 - t) n and -t n on b0 and b1, n the unit direction from q to p; an
active one (C < 0) is projected as a position-based constraint with
those gradients and the particles' inverse masses (Müller, Heidelberger,
Hennix and Ratcliff 2007, "Position based dynamics", section 3.3).
Friction is the position-based Coulomb model of Macklin, Müller,
Chentanez and Kim 2014 ("Unified particle physics for real-time
applications", ACM TOG 33(4):153, section 6.1), with one coefficient for
sticking and sliding: the relative tangential displacement of the two
closest points over the step is removed entirely while it is under mu
times the penetration depth, and shortened by mu times the depth above
it, distributed over the four particles as the normal correction is.
Candidate pairs come from a uniform grid whose cell is the longest segment
plus two radii, so two segments within two radii of each other have
midpoints in neighbouring cells; segments are visited in index order and
each one's candidates in ascending order, so the result is deterministic.
The grid describes positions at the start of this call and is not rebuilt
after each projection. Closest-point coincidence has no separating direction
and is skipped. The declared dt * referenceSpeed <= thickness admission
bound does not clamp actual particle speed; this discrete projection is not
continuous collision detection and cannot rule out tunnelling.
Project detected nonadjacent cord-segment penetrations along the line between their closest points, with Coulomb friction, in place on predicted positions. One ordered pass does not certify separation of every pair.
A cord is a one-wide lattice, its particles in index order along it and segment s joining particles s and s + 1. Each segment is a capsule of the domain's
thickness(its radius). For two segments, the closest points p = a0 + s (a1 - a0) and q = b0 + t (b1 - b0) come from the clamped segment-segment solution of Ericson ("Real-Time Collision Detection", 2005, section 5.1.9). The non-penetration constraint C = |p - q| - 2r (r the radius) has gradients (1 - s) n and s n on a0 and a1 and -(1 - t) n and -t n on b0 and b1, n the unit direction from q to p; an active one (C < 0) is projected as a position-based constraint with those gradients and the particles' inverse masses (Müller, Heidelberger, Hennix and Ratcliff 2007, "Position based dynamics", section 3.3).Friction is the position-based Coulomb model of Macklin, Müller, Chentanez and Kim 2014 ("Unified particle physics for real-time applications", ACM TOG 33(4):153, section 6.1), with one coefficient for sticking and sliding: the relative tangential displacement of the two closest points over the step is removed entirely while it is under mu times the penetration depth, and shortened by mu times the depth above it, distributed over the four particles as the normal correction is.
Candidate pairs come from a uniform grid whose cell is the longest segment plus two radii, so two segments within two radii of each other have midpoints in neighbouring cells; segments are visited in index order and each one's candidates in ascending order, so the result is deterministic. The grid describes positions at the start of this call and is not rebuilt after each projection. Closest-point coincidence has no separating direction and is skipped. The declared
dt * referenceSpeed <= thicknessadmission bound does not clamp actual particle speed; this discrete projection is not continuous collision detection and cannot rule out tunnelling.