[ODE] Russ' plans for ODE?

[Martin C. Martin]

gl wrote:

 > [Buoyancy]
>>My two cents: this really doesn't seem like ODE's place.  The tricky
>>thing about rigid body simulation is forces, like contact forces, that
>>can change a lot with a small change in position.  Buoyancy isn't really
>>in that category, same with rolling friction.
>>
>>I see ODE's roll as being a good core, rather than a swiss army knife of
>>everything someone might want to simulate.
> 
> Hmm, I disagree with that.  The things I mentioned are all related to
> physics, and would need to interoperate with the internals in any
> implementation - and are provided by other physics SDKs.

I disagree that it would need to interoperate with the internals, the 
only way it effects the simulation would be through an added force, so 
the buoyancy code would only need to call dBodyAddForce(), not any 
internals.

Also, ODE doesn't maintain the information needed to compute the 
buoyancy force.  ODE can currently handle objects with any distribution 
of mass that the user can compute the inertia matrix for.  Just knowing 
shape, inertia matrix, and total mass isn't enough to figure out how 
much mass is on one side of a plane.

Even if we assume uniform distribution for the buoyancy part, then the 
problem is largely geometrical: find out how much of an object is on one 
side of a plane.  To get the force, you simply multiply that by the 
gravity force, in the direction normal to the plane.

I think the difference is between an "engine for rigid body simulation" 
and a "physics SDK."  An engine is just the heart of a car, not a whole 
car.  I don't think ODE is trying to be a physics SDK.

Russ, are you trying to stay along the "rigid body engine" lines (doing 
the hard parts well), or are you thinking of evolving into a physics SDK 
(something that anyone can plop in their application that does all their 
physics needs?)

 >> What do you mean by convex hull collider ? How is that different from
 >> the Tri collider ?
 >
 > I believe two moving convex hulls can colide much more easily (and
 > reliably) than two moving trimeshes (and  should be more efficient).

I think a more proper term for this would be "convex tri meshes."

- Martin