JuliaSimControl migrated to Dyad

[stepanoslejsek]

The homework (https://hurak.github.io/orr/rocond_hw.html) related to robust control is based on JuliaSimControl package, which is no longer available since the migration to Dyad.

Writing this issue as a sort of TODO for myself.

[baggepinnen]

Are you sure you need DyadControl or would hinfsynthesize from RobustAndOptimalControl.jl suffice?

[hurak]

I confess that I have never really studied/understood the role of JuliaSim within the control-oriented computing, and when our last year TA @uchytil decided to use JuliaSim instead of just RobustAndOptimalControj.jl, I let him do as he wished last year. Do you, @uchytil still remember the reason for choosing so? And could you, @stepanoslejsek, try to explore if the particular homework problem could rely on just RobustAndOptimalControj.jl? It would be certainly great not to force students to create an account anywhere.

[hurak]

I think I start recalling the reason for Adam's preference for JuliaSIm(Control) – isn't it the case that it is only their implementation that also offers to choose the LMI-based solver that is free of those "troubles on the imaginary axis"?

[baggepinnen]

Yes, RobustAndOptimalControl only implements the method that solves two Riccati equations, DyadControl has a number of LMI methods for Hinf and related problems.

[uchytil]

I believe the problem was that hinfsynthesize from RobustAndOptimalControl.jl requires the nonsingularity of the feedthrough matrices D12 and D21 of the generalized plant. We were not able to satisfy this in our use case.

[stepanoslejsek]

It is precisely as @uchytil said. Yesterday, I tried to use only RobustAndOptimalControl.jl for that particular homework problem, and I ran into the same issue. Below is the error message.

ERROR: Cannot scale the system, assumption A2 is violated. This typically means that the control input is not penalized by D12 or that there is missing measurement noise in D21. Stacktrace: [1] error(s::String) @ Base .\error.jl:35 [2] #_scalematrix#127 @ C:\Users\oslejsek\.julia\packages\RobustAndOptimalControl\tJZIs\src\hinfinity_design.jl:566 [inlined] [3] _scalematrix @ C:\Users\oslejsek\.julia\packages\RobustAndOptimalControl\tJZIs\src\hinfinity_design.jl:562 [inlined] [4] _transformp2pbar(P::ExtendedStateSpace{…}; kwargs::@Kwargs{…}) @ RobustAndOptimalControl C:\Users\oslejsek\.julia\packages\RobustAndOptimalControl\tJZIs\src\hinfinity_design.jl:538 [5] hinfsynthesize(P::ExtendedStateSpace{…}; gtol::Float64, interval::Tuple{…}, verbose::Bool, tolerance::Float64, γrel::Float64, transform::Bool, ftype::Type, check::Bool, method::Symbol) @ RobustAndOptimalControl C:\Users\oslejsek\.julia\packages\RobustAndOptimalControl\tJZIs\src\hinfinity_design.jl:158 [6] top-level scope @ REPL[118]:1 Some type information was truncated. Use show(err) to see complete types.

[hurak]

Yes, that is the well-known limitation of the Riccati equation-based solvers for the Hinf optimal control problem...