The sparta code keeps track of a number of structures for computation. These are:
POINTS: Basically a vertex with some
inherent mass.
SPRINGS: In order to tweak the velocity of
two connected points according to their
relative position. Springs can be set to
break if their lenght exceeds some
threshold. We call stiff, breakable,
inter-object springs virtual staples.
FACES: Made up of 3 points. They are for collision detection and display updates.
CDLINES: These are line segments between 2 points for collision detection purposes. Originally we used springs, but this is more efficient, and avoids some friction problems.
Here is some simplified psuedocode:
For each step
{
For every object
For every spring
{
compute force due to displacement
compute force due to damping
}
For every object <- There are other methods with better order
For every object but our cpus limit us to small N anyhow
For every cdline in object 1
For every face in object 2
{
Do bounding box check for early exit
if we have a collision (this is complicated)
tweak velocities taking into account mass
and friction
}
}
For each object
For each face
draw triangle
Collision detection: we picked an unconventional collision detection approach since some more traditional approaches (inside/outside polygon test) have problems with flexible objects. For example, it is impossible to guarantee that a flexible object is going to stay convex.
Rigid bodies: Some performance gains could be made by limiting these simulations to rigid bodies. Inside/outside polygon test would work, and standard 3d hardware could be used to accelerate object transformations. In the SPARTA project, we are interested in flexible objects, however.