[ODE] Infinite sliding... ?

[Jon Watte (ODE)]

Jeff Kershner wrote:

> Second:  This is the third time I have heard that stepping at variable times
> is very wrong.  I can't find the FAQ about this except that it is wrong.  In
> a normal game loop that is operation almost all the time at 75 Hz, how much
> should the step size be and how often should it be called every frame?

If you know you run at 75 Hz, then set step size to 1/75. For the cases 
where you drop frames, step more than once. An alternative is to set 
step size to, say, 1/150, and step 2 or 3 times depending on frame 
duration. Over time, the accumulated physics time should be the same as 
the accumulated frame time.

The reason non-equal step size is bad, is that objects keep falling into 
ground (because of gravity), but "stabilize" at a small depth into the 
ground. When step sizes vary, the ideal "stable" position will change 
each step, and thus the object will jitter (and may very well gain energy!)

> Third: can't do this since I need to interact with the bodies after they lay
> at rest.

You can re-enable them when you want to interact with them. In the 
collision callback function, for example, if you collide with a 
non-suspended body.


> This is what I do:
> const dReal* v = dBodyGetLinearVel(it->body);
> dBodyAddForce(it->body, v[0] * -0.01, v[1] * -0.01, v[2] * -0.01);

Looks good.

> // Also so this for angular vel with a coefficient greater than 1
> const dReal* av = dBodyGetLinearVel(it->body);
> dBodyAddForce(it->body, av[0] * 2.0, av[1] * 2.0, av[2] * 2.0);

Lookas all wrong. First, it uses the linear velocity. Second, it ADDS to 
the velocity (because 2.0 is not negative). Third, it still uses force. 
You probably want to write this as:

   dReal const * av = dBodyGetAngularVel(it->body);
   dBodyAddTorque(it->body, av[0] * -0.03, av[1] * -0.03, av[2] * -0.03);

However, you really DO need to enable auto-sleeping. That's how you get 
things to "stick" rock solid, in any simulator I've seen.

Cheers,

			/ h+