CoM Trajectory Generation from HMC

[arntanguy]

This PR is an ongoing implementation of some of HMC's CoM generation features, mainly the long-term and short-term CoM trajectory generation introduced in

Online 3D CoM Trajectory Generation for Multi-Contact Locomotion Synchronizing Contact, Mitsuharu Morisawa et al., IROS 2018

In its current state it is not necessarily meant to be merged, but mostly to show that the current status of this implementation should someone wish to take over (@mmurooka, @mitsuharu-morisawa). It might be worth splitting this work into its own mc_planning project, and merging it with @mmurooka's own planning developments of #109.

• For a minimalist usage code sample, please refer to mc_sample_CoMGeneration (code / documentation). It is roughly equivalent to hmc's generator example

• This can be tested using the new sample controller CoMTrajectoryGeneration (code). It uses the same generator class used by mc_sample_CoMGeneration to generate a CoM trajectory between footsteps, and tracks it using LIPMStabilizer. You can run this sample controller with

MainRobot: HRP5P # or HRP2DRC, HRP4, etc
Timestep: 0.002
Enabled: CoMTrajectoryGeneration

Here is a video of what to expect:

vokoscreen-2021-03-30_16-56-52.mp4

The generated documentation for this branch is available at https://arntanguy.github.io/mc_rtc/doxygen.html:

• The main changes are implemented in the mc_planning namespace.
• The most relevant class to look at is the generator class that implements an easy to use .

In addition this PR implements:

• A generic LookupTable implementation (code / documentation) to pre-compute computationally expensive functions (in this case cosh/sinh that are called for the whole time-horizon at every iteration

• A basic tool mc_bin_compare (81f44d8) to compare data between log files. This is used here to check for regressions (
  9b1bd15) against a reference log file of what the generator sample mc_sample_CoMGeneration should do.

• Utilities to safely iterate over preview windows and convert between time/indices. The intent was to limit the risk of mistakes in handling indices when iterating over preview windows. You can see for instance the PreviewWindowView

Missing:

• feeding back the real robot state into the CoM trajectory generation. I believe that implementing this would provide a significant improvement in the quality of stabilization in mc_rtc, by no longer only relying on the stabilizer's tracking performance, but also adapting the reference trajectory to account for perturbations.
• Multi-contact implementation from

  Multi-contact Stabilization of a Humanoid Robot for Realizing Dynamic Contact Transitions on Non-coplanar Surfaces - Morisawa et al.

• walking example