Add filter method alternative to linesearch

[edantec]

This PR:

• Adds a pair filter option to manage step acceptance in place of linesearch
• Provides a python example of a walking OCP on Talos.

Resolves #31

[ManifoldFR]

Thanks ! I'll do some editing to this PR if you don't mind

[edantec]

In commit [a7dee9f] I point out a circular include of enums.hpp in solver-proxddp.hpp

[jcarpent]

In commit [a7dee9f] I point out a circular include of enums.hpp in solver-proxddp.hpp

It should not be a problem is there is the classic safe guard or the new pragma once.
Is it the case?

[ManifoldFR]

I have a few comments

[ManifoldFR]

The CI - OSX/Linux via Conda build is failing because we're using the features for std::pair from the new eigenpy 3.2.0 and the conda recipe is still on the previous version of eigenpy. Switching to std::array<_, 2> also won't do it for the same reason

We can do three things:

• upgrade the eigenpy requirement to 3.2.0 for the entire library (fine by me) and the conda recipe
• replace std::pair<Scalar, Scalar> by a simple struct
• manually expose the pair type (with an eigenpy::register_symbolic_link safeguard in case another library chooses to expose the same type)

The first two options seem like best practice to me.

@jorisv what do you think ?

[ManifoldFR]

By the way @edantec please add yourself as a contributor to the README 😃

[jcarpent]

The CI - OSX/Linux via Conda build is failing because we're using the features for std::pair from the new eigenpy 3.2.0 and the conda recipe is still on the previous version of eigenpy

A new version of Pinocchio with the latest eigenpy release is on its way (see conda-forge/pinocchio-feedstock#107).

[ManifoldFR]

I ended up bumping the required eigenpy version to 3.2.0.

[jcarpent]

@edantec What is the status of this PR?

[edantec]

@edantec What is the status of this PR?

From my side it is clean and finished. We are waiting for Wilson and Joris to fix the CI and make relevant updates on proxsuite-nlp.

[ManifoldFR]

New packages for proxsuite-nlp 0.3.2 (using eigenpy 3.2.0) got shipped, I'll trigger the CI jobs again.

[ManifoldFR]

@edantec Please update the changelog

[edantec]

There are two tests that randomly fail on macos and I have no idea why (probably memory leak somewhere).
First failing test in python/test_solver.py where a basic LQR solved with proxddp linesearch does not converge sometimes.
Second failing test in python/test_integrators.py where the finite differences of RK2 integrator are wrong (in forward explicit dynamics).

I managed to reproduce the results on a Macos system, but I struggle to find the causes as I'm not used to debug under Macos. Some help would be welcomed!

[ManifoldFR]

Oh wow, I never noticed this.

[edantec]

The memory error on test_solver.py has been fixed. As for the memory error of test_integrators.py, I have no idea why but it seems to have disappeared according to Joris.

[ManifoldFR]

The memory error on test_solver.py has been fixed. As for the memory error of test_integrators.py, I have no idea why but it seems to have disappeared according to Joris.

It's the magic of ChristmasC++

[ManifoldFR]

Everything looks copacetic. Will make a PR later to upstream the filter methods into proxsuite-nlp and remove the exceptions.