next »

[zarovv]

Hi.
I asked in the past about an issue i was having with collision and i was suggested to try out impulse based collision response.

I tried it and it works pretty okayish, but i have some annoying problem of collision jitter.

Basically, when i push the cube in down direction it jitters right and left non-stop.
It just rotate right,left,right and doesn't seem to slow down, stop and align with the ground.

Someone knows or have a clue on how to handle it?




Sorry that i bother with a lot of questions. its just this community seems to be responsive and very helpful )

[qMopey]

Your best bet is to learn from a resource that already works. For example, Box2D Lite (this is the Lite version, not actual Box2D, it's just a tiny demo to learn from). That way you can compare your code to it. Box2D Lite has a sophisticated solver that solves multiple collision points simultaneously to avoid jitter. https://github.com/erincatto/box2d-lite

If you want something even easier to understand, without the sophisticated solver, just some naive impulses, there is also this one: https://github.com/RandyGaul/ImpulseEngine

[buto]

Maybe a little late... just wanted to say that the demo looks super cool!

[zarovv]

If someone is interested... after many attempts, crying and frustration i managed to make it work:

I did it in 3d actually and the main algorithm (Is for 2 colliding objects. A is moving B is stationary and will always be) is:

   Vector rA = contact - center;
   Vector vr = linearVelocity + CROSS(angularVelocity, rA);

   float contactVel = DOT(vr, normal)
   ;

   Vector raCn = CROSS(rA, normal);

   float raI = DOT(raCn, VEC_MULT(invertedInertia, raCn))
   ;

   float lambda = (-contactVel + pushDistInNormalDirection)
         / (1.0f / mass + raI);

   Vector impulse = VEC_MULT_SCALAR(lambda, normal);

   Vector linearVelocityDiff = VEC_MULT_SCALAR(1.0f / mass, impulse) / contactCount;
   Vector radiansVelocityDiff = VEC_MULT(invertedInertia,
         CROSS(rA, impulse)) / contactCount;

Search for Randy Gaul 3D. He has some simple enough impulse based git project also...

[qMopey]

I'm Randy Gaul, and I posted in this topic a while ago

If anyone has questions about this kind of stuff I'm easy to contact on twitter.

[zarovv]

I'm Randy Gaul, and I posted in this topic a while ago

If anyone has questions about this kind of stuff I'm easy to contact on twitter.

OOOOOOoooo  
A legend indeed. Your implementation is the easiest I found online.
I struggled for example with the coefficient for the impulse, always used a constant coefficient.
But then looking at your implementation I tried using the push distance(penetration) as the coeff and it worked.
Thanks )