Roadmap March 2023

[ocots]

Starting tasks:

• @jbcaillau @joseph-gergaud @PierreMartinon Box and pure constraints for the direct method.
• @jbcaillau @ocots Getters of multipliers associated to labeled constraints.
• @ocots @joseph-gergaud @PierreMartinon Functional: Bolza cost.
• @joseph-gergaud @PierreMartinon Direct solve by Crank-Nicolson.
• @joseph-gergaud @PierreMartinon Direct solve by generic RK scheme.
• @joseph-gergaud @PierreMartinon Sparse usage (Jacobian, Hessian of the constraints).
• @PierreMartinon Julia wrapper of bocop for benchmarking.
• @ocots @PierreMartinon Common optimal control solution with plot for the direct methods.
• @ocots @PierreMartinon Common optimal control solution with init for the direct methods.
• @ocots improve solution (control) flow plot.
• @ocots @joseph-gergaud Ipopt with the direct shooting method.
• @ocots @PierreMartinon Replace the print callback of ipopt.

You can get access to the list of all the issues labeled march2023.

Remarks / other possible tasks.

• we have to start the documentation!
• we could add an indirect shooting method: see Issue 24.
• for the direct shooting method, we could generalize the problem solved by the unconstrained solver (need to generalize also the adjoint method for better performance). See Issue 1. We have to add the option of the state discretization: see also Issue 2 and Issue 9.
• benchmark of the direct shooting method with and without adjoint method
• usage of assert macro for pre-conditions: see Issue 74.
• Dynamics: output and in-place
• we need to add problems in CTProblems starting by adding the solution of the Goddard problem.

[ocots]

We must handle print of OptimalControlSolution from CTBase.jl:

> sol = solve(ocp)
Error showing value of type OptimalControlSolution:
ERROR: MethodError: no method matching get(::IOContext{Base.TTY}, ::Symbol, ::Bool)
You may have intended to import Base.get

Some of the types have been truncated in the stacktrace for improved reading. To emit complete information
in the stack trace, evaluate `TruncatedStacktraces.VERBOSE[] = true` and re-run the code.

Closest candidates are:
...

[ocots]

I have added for the moment a method to Base.show in OptimalControl.jl (that should be moved to CTBase.jl.

import Base: show, Base
# we get an error when a solution is printed so I add this function
# which has to be put in the package CTBase and has to be completed
function Base.show(io::IO, ::MIME"text/plain", sol::OptimalControlSolution)
    nothing
end

[ocots]

When testing OptimaControl.jl I have the following message:

WARNING: both CTBase and Base export "adjoint"; uses of it in module OptimalControl must be qualified
WARNING: both CTBase and Base export "success"; uses of it in module OptimalControl must be qualified

Maybe, in CTBase.jl, add

import Base: adjoint, success

to extend the functions.

[ocots]

I am not really sure that getters is a good thing. It is quite simple and nice to access directly the field of struct. In that case, there is no need of adjoint and success methods in CTBase.jl.

[ocots]

I have also this message:

WARNING: importing deprecated binding Colors.RGB1 into PlotUtils.
WARNING: importing deprecated binding Colors.RGB4 into PlotUtils.

[ocots]

About Typing, maybe I should remove the Function Type in:

• https://github.com/control-toolbox/CTBase.jl/blob/main/src/macros.jl#L15
• https://github.com/control-toolbox/CTBase.jl/blob/main/src/solutions.jl#L29

since for instance, when using Interpolations.jl the interpolation function is not a subtype of Function (which is sad).

[ocots]

Coding Jupyter notebook and github: https://docs.github.com/en/repositories/working-with-files/using-files/working-with-non-code-files#working-with-jupyter-notebook-files-on-github

[ocots]

We can enable rich diff for jupyter notebooks:

[PierreMartinon]

Nice !

[ocots]

@jbcaillau I have made a new release of OptimalControl (v0.3.0). See the notebooks in the examples folder.

[ocots]

@PierreMartinon I have removed internal getters, that is getters of for instance the DirectSolution since this struct is internal to CTDirect. Defining getters produce more functions and we have to be careful since the package Base has already functions like adjoint or success.

[PierreMartinon]

Well if the getter was just returning the field, we can choose direct access yes. We can still reintroduce getters later if we decide so at some point.

[jbcaillau]

Issue in box constraints transferred to CTBase.jl. Some other issues should be moved... (will almost surely break some links, though)

[ocots]

I had put all the issues in OptimalControl.jl since the labels (or tags) are by repository. But no problem, I will add a label march2023 in CTBase.jl.

[ocots]

We have to be better on the covering, that is we need more unit tests.

[ocots]

I have replaced CTProblemLibrary by CTProblems and HamitlonianFlows by CTFlows.

Now, we have to update any package which need it, that is update the Project.toml.

julia> ]
rm CTProblemLibrary HamitlonianFlows
add CTProblems CTFlows