NewtonUpdate timestep value

[misho]

http://gamedev.stackexchange.com/questions/1589/when-should-i-use-a-fixed-or-variable-time-step

Interesting. If I understand this, the way to go is to let Physics loop run at one "speed", and have graphics engine "fetch" results at every frame update. There could be many Physics updates between every frame update... makes sense?

[JoeJ]

Code: Select all

void NewtonUpdate(const NewtonWorld* const newtonWorld, dFloat timestep)
{
   TRACE_FUNCTION(__FUNCTION__);
   Newton* const world = (Newton *)newtonWorld;

   dgFloat32 minstep = dgFloat32 (DG_MIN_TIMESTEP);
   dgFloat32 maxstep = dgFloat32 (DG_MAX_TIMESTEP);

   timestep = dgClamp (dgFloat32 (timestep), minstep, maxstep);

//NewtonSerializeToFile (newtonWorld, "xxx.bin", NULL, NULL);
   world->UpdatePhysics (timestep);
}


You can change min/max timestep defines and see if the simulation still runs fine.
For a flight sim this might be ok, because there are no constraints and mostly just integration.

Otherewise fixed timestep is recommended for stable simulation, which means async graphics and physics and you need to interpolate current und previous physics state for graphics to keep graphics looking smooth.

[Windrider]

Could you please elaborate a little on "you need to interpolate current und previous physics state for graphics to keep graphics looking smooth" please?

[misho]

I always thought this line (from tutorials) took care of the interpolation:

Code: Select all

   // calculate the interpolation parameter for smooth rendering
   g_sceneManager->SetIntepolationParam(dFloat (g_timeAccumulator) / dFloat(DEMO_FPS_IN_MICROSECUNDS));

along with

Code: Select all

   dVector posit (m_prevPosition + (m_curPosition - m_prevPosition).Scale (interpolationParam));
   dQuaternion rotation (m_prevRotation.Slerp(m_curRotation, interpolationParam));
   m_matrix = dMatrix (rotation, posit);

Is that correct?

[JoeJ]

Yes, this seems the code to handle interpolation for demo sandbox.
It's also described in that famous article http://gafferongames.com/game-physics/fix-your-timestep/

[misho]

Thanks for confirming that! Looks like a cozy bedtime read!

[Sweenie]

I usually do something like this...

Code: Select all

const float TimeStep = 1.0f / 60.0f; // 60 physics updates per second

float accumulator = 0.0f;

void PhysicsUpdate(float deltaTime)
{
  accumulator += deltaTime;
  while (accumulator >= TimeStep)
  {
    NewtonUpdate(TimeStep); // A step of 0.016 seconds will occur when 0.016+ seconds has passed.
    accumulator -= TimeStep;
  }
}

Now, you should be able to implement a timescale by doing something like this...

Code: Select all

const float TimeStep = 1.0f / 60.0f; // 60 physics updates per second
const float TimeScale = 2.0f;
float accumulator = 0.0f;

void PhysicsUpdate(float deltaTime)
{
  accumulator += deltaTime;
  while (accumulator >= TimeStep * TimeScale)
  {
    NewtonUpdate(TimeStep); // A step of 0.016 seconds will occur when 0.032+ seconds has passed.
    accumulator -= (TimeStep * TimeScale);
  }
}

The above change should make the simulation go at half speed while still stepping the simulation at at fixed 0.016 second time steps.

But objects will not move smoothly when slowing down time but choppy so if you want "Matrix" style slowmotion you will probably need to implement some interpolation tweening as well. Gaffer on Games explains it pretty well as Joe mentioned.