Internal representation of optimization variables about ifopt HOT 8 CLOSED

When you create a problem, you assign variable sets like what is shown below with unique names. Then when you add a constraint or cost to the problem it gets linked to the problems variable set. So inside the cost or constraint object you have access to the member function GetVariables() which gives you access to all variable sets. Then during the solve it will call the GetValues() and FillJacobianBlock. I also included an example of a constraint implementation which may help.

  Problem nlp;
  nlp.AddVariableSet(std::make_shared<ExVariables>("var_set0"));
  nlp.AddVariableSet(std::make_shared<ExVariables>("var_set1"));

Example:
https://github.com/tesseract-robotics/trajopt/blob/master/trajopt_ifopt/include/trajopt_ifopt/constraints/joint_velocity_constraint.h
https://github.com/tesseract-robotics/trajopt/blob/master/trajopt_ifopt/src/constraints/joint_velocity_constraint.cpp

from ifopt.

Thank you @Levi-Armstrong. But, as I understand, you are getting the values of the external representation (as a VectorXd). I was hoping I could get the internal representation mentioned in the following code snippet. I'd like to define some x0_ and x1_ as Fourier series, for example, and make use of them inside the Constraint and Cost classes.

// Here is where you can transform the Eigen::Vector into whatever
// internal representation of your variables you have (here two doubles, but
// can also be complex classes such as splines, etc..
void SetVariables(const VectorXd& x) override
{
x0_ = x(0);
x1_ = x(1);
};

from ifopt.

You need to create your own variable set by inheriting from VariableSet like what is shown in this example.

Are you wanting to know how to access the x0_ and x1_ from your cost or constraint function?

from ifopt.

Yes, I already have a variable set, and I wanted to access x0_ and x1_ (two Fourier series in my case) from the constraint set. If it isn't possible, then I fail to see the overall usefulness of x0_ and x1_

from ifopt.

Add two methods to your variable set class getX0() and getX1() then in your cost or constraint.

auto my_set = std::dynamic_pointer_cast<MyVariableType>(GetVariables()->GetComponent("var_set1"));
... now call your methods exposing x0_ and x1_

from ifopt.

Thank you, I'll try it out.

from ifopt.

Add two methods to your variable set class getX0() and getX1() then in your cost or constraint.

auto my_set = std::dynamic_pointer_cast<MyVariableType>(GetVariables()->GetComponent("var_set1"));
... now call your methods exposing x0_ and x1_

Thank you very much. It worked like magic. I hadn't heard about dynamic_pointer_cast. I was afraid the solution was something much simpler that I wasn't seeing.

from ifopt.

Yea, I think this is the only option you have if you want implementation specific access.

from ifopt.