Prescribed motion

[Michael Sherman]

One other unrelated comment:

The VerletIntegrator is almost always a bad choice for a multibody system. I would suggest that you use RungeKuttaMersonIntegrator instead for most purposes.

Regards, Sherm

[Kishor Bhalerao]

Hi Sherm,

Just to follow up on this discussion. Right now, we have got the prescribed motion working with the RNA problem in Simbody and the system seems to behave as expected. I am now attempting to use it in my DCA solver.

In main(), after defining the PrescribedMotion (this is a function of time), I expected that on calling the function system.realize(state,Stage::Velocity), the state of the system would be updated to reflect the changes due to the prescribed motion. However this does not seem to be the case. Here is what I have been doing.

// defined on MobilizerQIndex(4)
Constraint::PrescribedMotion con_1(....)

// defined on MobilizerQIndex(5)
Constraint::PrescribedMotion con_2(....)

system.realizeTopology();
State& state = system.updDefaultState();
cout<<state.getQ();

// Change the value of time without changing anything else in the system
// I expected this to change the values of both Q(4) and Q(5)
state.updTime()=10.0;

system.realize(state,Stage::Velocity);
cout<<state.getQ();

The Q in both the cases is the same. I am not sure why this is the case. The prescribed motion function defined in con_1 and con_2 is definitely being called and the values generated by calcValue() and calcDerivative() are exactly what I expect them to be. But these do not get reflected in the state.

Do you have any suggestions ?

Thanks,
Kishor.

PS: The search feature in the doxygen docs for simtk2.0 is very useful !

[Michael Sherman]

Hi, Kishor.

Prescribed motion in Simbody is currently treated exactly like any other constraint. Realizing calculates the constraint *errors*; it does not enforce the constraints (except acceleration constraints which are used to calculate the udots and multipliers).

The realize(state,stage) method takes the state as a "const" value -- it can't make state changes. For state changes you need what Simbody calls a "solver", meaning some method that takes a writable state. The solver you are looking for in this case is system.project() which takes a state that violates position and/or velocity constraints and projects it onto the position and velocity constraint manifolds. This is the solver used by the integrators' initialize() methods prior to the start of a simulation.

Unlike "real" prescribed motion (which I'm working on slowly) the Constraint::PrescribedMotion has to fight equally with other constraints in the system, so it is subject to the usual assembly and velocity analysis caveats: you might have to be reasonably close to a solution to get there, e.g.

project() takes several odd arguments; you might find it easier to create an integrator and call initialize() since the integrator will know how to set up the call.

Regards,
Sherm

[Kishor Bhalerao]

Hi Sherm,

This clearly explains the results i was getting from my code. I suppose, in the current example, the easiest way for me would be to simply use state.setQ() and state.setU() directly. I should perhaps not include the Prescribed Motion constraint on the system at all since, i do not expect to realize the system to acceleration stage using simbody methods.

Thanks for your help.
Kishor.