[ODE] OT: Physics engine history!

[Pierre Terdiman]

>   I also get a big one : my boss
>    believe that writing a physics engine like Havok is easy ???? sigh

It reminds me of a reunion, some years ago, where the art director basically
said something like :

"Collisions ? What collisions ? That's the simplest thing to do!"

Then colliding his two hands representing two objects :

"You get that object and that other one, and, well.... that's it, they
collide! There's a wall, you stop the character."

(...big silence in the room)

He was thinking collisions (and response!) were handled "automatically" in
some magic way....

(yes, I left this company with no regrets)

---

I don't know about the "physics engine history" but you can point out that :
- Havok bought Ipion and merged the technologies they both had been
implementing for years
- If you follow the Baraff SIGGRAPH papers you can see how it took him years
to refine his models / methods
- It's not a closed research topic, many papers dealing with rigid body
simulation are published all the time
- Even the contact manifold generation problem is an unsolved problem.
That's not a physics engine, that's a -tiny- part of a physics engine. Yet,
here's what Chris Hecker wrote once on the GD-Algo list (below)

etc, etc

Pierre

---

>I'm currently writing a physics simulator and suffer from the problem of
>contact determination. ...
>But I wasn't able to use the algorithm to compute multiple contact points.

This is an incredibly hard problem, and unfortunately, it's also an unsolved
incredibly hard problem.  Last year's phyics seminar was totally dedicated
to this topic, and the consensus from all the researchers and attendees
after two 10-hour days is that there's still no algorithm that works very
well.  :)

Your best bet would be to find somebody who has a copy of the proceedings
from the physics seminar from last year, since it has all the relevant
papers in it, including some that weren't published anywhere else.
Unfortunately, we don't have any left.  The researchers you're going to want
to look for are Vanecek, Bouma, Jing Xiao, Baraff (he hints about his
algorithm in his contact papers), and Cremer.  A trip through CiteSeer
should help.  Xiao is the only person working on this right now, and I don't
think even she's doing it very actively.  The algorithms range from
iterative hacks to attempts at solving the full regularized boolean set
operations.  Baraff's algorithm (used by Maya) is a little unclear, but he
says he's just examining local areas and trying to figure things out that
way.  His works for concave objects.

We're going to talk a bit about calculating manifolds at the seminar this
year.  It's mostly going to be manifolds between soft bodies and rigid
bodies (since the focus is on fluids and soft bodies for special effects).
Obviously soft bodies makes it even harder (there's a paper by Snyder on
doing this with lots of interval arithmetic).

Chris