Step by step integration

[Andrew Stolin]

Hi,

I am trying to perform a simulation step by step in python. The results differ between a 0.2s integration and two 0.1s integrations.

Would you be able to tell me if there is something wrong in these few lines of code?

I am using the build_simple_arm_model.py" example.

The controller

Code: Select all

brain = osim.PrescribedController()
brain.addActuator(biceps)
func = osim.Constant(1.0)
brain.prescribeControlForActuator("biceps", func)

1 step (0.2)

Code: Select all

manager = osim.Manager(arm)
state.setTime(0)
manager.initialize(state)
state = manager.integrate(0.2)

The check

Code: Select all

print(elbow.getCoordinate().getValue(state))
print(biceps.getActivation(arm.initSystem()))
print(biceps.getExcitation(arm.initSystem()))

gives me :
1.5007997946757727
1.0
0.9999787865083376

2 steps (0.1 then 0.1)

Code: Select all

manager = osim.Manager(arm)
state.setTime(0)
manager.initialize(state)
state = manager.integrate(0.1)
state.setTime(0)
state = manager.integrate(0.1) # I believe that the manager.integrate is always based on the current state

The check
gives me :
1.5007492993654519
1.0
0.999978786034814

Any idea?

[Thomas Uchida]

The results differ between a 0.2s integration and two 0.1s integrations.

This behavior is expected. The Manager uses a SimTK::RungeKuttaMersonIntegrator, which is a variable-time-step integrator. The differences you're seeing are about 5e-5 and 5e-10, which are likely below the integrator's default tolerance and would be explained by the integrator choosing different time step sizes depending on whether you're forcing it to stop exactly half-way through.

[Andrew Stolin]

Thank you very much for this information, I did not think of the integrator's default tolerance.

Just to confirm, does it mean that when I re-integrate, all the controls have been preserved by default (i.e. I am integrating from the last state of the last integration, and not from the beginning) ?

Code: Select all

manager = osim.Manager(arm)
manager.initialize(state)
arm.initializeState()
state.setTime(0)
state = manager.integrate(3.0)

# second integration
arm.initializeState()
state.setTime(0)
state = manager.integrate(3.0)

As a side note, I am not sure why we have to setTime to 0 every time in order to do more than two successive integrations.

[Thomas Uchida]

does it mean that when I re-integrate, all the controls have been preserved by default

Not sure, but it seems like it would be relatively easy to test.

I am not sure why we have to setTime to 0 every time in order to do more than two successive integrations.

Does Manager::resetTimeAndDTArrays() not work? You might also want to check TestManager.cpp for examples (https://github.com/opensim-org/opensim- ... anager.cpp).

[Hide Kimpara]

Hello Andrew,

manager.initialize(state)
state = manager.integrate(0.1)
state.setTime(0)
state = manager.integrate(0.1) # I believe that the manager.integrate is always based on the current state

As you thinking, manager.integrate starts from current state time. Therefore, this mentioned code looks like twice integration of 0 >> 0.1 [s].
If you are willing to 2 step integrations such as 0.0 >> 0.1 [s], and 0.1 >> 0.2 [s], you may do integrations as following:

Code: Select all

state.setTime(0)
manager.initialize(state)
state = manager.integrate(0.1)
state = manager.integrate(0.2)

if you are willing to see the time change in opensim's state variable, you can use "state.getTime()" just after integration.

I hope this would be help for you.

Hide

[Andrew Stolin]

Thank you very much, Thomas and Hide - everything seems to be working now