Cost Model Force Linear Cone

This creates a cost model which implements -> residue = A*f-ref. The idea is to use this residue and the activationmodelinequality to get Af<=0.

I have implemented a unittest for this cost:

• I have only implemented the Lx and Lu checking, since Lxx=rx*rx does not apply in this case.
• I have only implemented a simple checking of the derivatives, since activationmodelinequality is already being checked. So I don't check here which inequality is active.

This PR is related to #116 (closed).

mentioned in issue #116 (closed)

changed the description

Summary of the oral discussion: yes it should be A.TAxxA, and not A.TAxx.TAxx*A. In the specific case where Axx is a diagonal of binary elements, it is equivalent, but might lead to errors when other activations are used.

As we are only planning to use inequality activations, let's make this code more specific and avoid the expensive A.T*Axx muliplication: this should be replace by a specific operations selecting the nonzero rows for a smaller multiplication.

marked as a Work In Progress

added on-going label

assigned to @proyan

added 33 commits

• 45c5b16e...2b4c4927 - 30 commits from branch loco-3d:devel
• 54ec0f91 - [ForceLinearCone] create a cost which takes implements Af<=0, where A is a...
• ccfb29aa - [unittest] unittest for linear cone
• 7a576f24 - [LinearCone] fix Axx, and create assertion to ensure activation is always inequality

Compare with previous version

resolved all discussions

added 1 commit

• 726b2a20 - [LinearCone] use selection instead of matrix mult for A.T*Axx*A

Compare with previous version

unmarked as a Work In Progress

@nmansard This should be ok to merge now.

removed on-going label

You can merge after solving the first comment (the second one is an option).

added 7 commits

• 726b2a20...c67ebebc - 2 commits from branch loco-3d:devel
• 2bd11944 - [ForceLinearCone] create a cost which takes implements Af<=0, where A is a...
• b9f0c8c1 - [unittest] unittest for linear cone
• 5924c5e3 - [LinearCone] fix Axx, and create assertion to ensure activation is always inequality
• 567b7818 - [LinearCone] use selection instead of matrix mult for A.T*Axx*A
• c7173e5a - [Linear Cost] review comments on !123 (merged): 1) don't allow user to set activation...

Compare with previous version

resolved all discussions