Move Dynamical Inverse Kinematics to HumanDynamicsEstimationLibrary

[lrapetti]

This PR moves the Dynamical Inverse Kinematics implementation from the HumanStateProvider device to the HumanDynamicsEstimationLibrary.

With this PR all the functionalities of the current implementation are ported to the new class replicating the behaviour.

While addressing this refactoring some relevant points and possible improvements have been detected, and some of them addressed:

• [x] Dynamical Inverse Kinematics naming has been preferred to Integration-based Inverse kinematics (accordingly to reference work https://www.mdpi.com/1999-4893/13/10/266). Indeed any configuration file out of this repo should be updated (possibly we should use deprecation rather then changing directly the name in this PR, but not sure if those configuration files are used in many place rather then inside this repo itself)
• [ ] Interface for setting velocity limits should be improved, I have replicated those of InverseVelocityKinematics class, but all the limits should be sent together.
• [x] I have verified that https://github.com/robotology/human-dynamics-estimation/issues/167 is happening if setCustomConstraintsJointsValues is not called. While addressing point above we should fix also this
• [ ] We should implement a test. We can use https://github.com/robotology/idyntree/blob/master/src/inverse-kinematics/tests/InverseKinematicsUnitTest.cpp as reference
• [ ] Currently adding/removin targets in real-time is not possible https://github.com/robotology/human-dynamics-estimation/issues/267
• [ ] Currently multiple target for the same link can not be used https://github.com/robotology/human-dynamics-estimation/issues/266
• [ ] Currently 3D position targets are only implemented, we should add the possibility to constraint only one of the three component
• [ ] Integral error correction term is currently missing
• [ ] Kinematic constraints handling should be moved from inverse-velocity-kinematics class to dynamical-inverse-kinematics

Before merging

• [x] fix the gains/weights handling

[S-Dafarra]

Just to understand, what is the difference between Dynamical Inverse Kinematics and Inverse Velocity Kinematics?

[lrapetti]

Just to understand, what is the difference between Dynamical Inverse Kinematics and Inverse Velocity Kinematics?

Dynamical Inverse Kinematics class is solving the IK problem. Inverse Velocity Kinematics instead basically computes the configuration velocity $\nu$ solving the differential kinematics $\mathrm{v}=J\nu$. Indeed, dynamical inverse kinematics uses the inverse velocity kinematics solver for computing the velocity to be integrated.

[lrapetti]

Just to mention, I took the change also to do some cleaning in HumanStateProvider:

• deprecate useXsensJointAngles option that has never been made completely working, and never used (https://github.com/robotology/human-dynamics-estimation/pull/265/commits/80d8c7af0785c2278f9b835edb28f943138fe4b9)
• remove some unused variables https://github.com/robotology/human-dynamics-estimation/pull/265/commits/1c644cefb8fd7c6b0fe7fa1dc1f482ef62e965f5 (in particular integral term for dynamical inverse kinematics can be re-added once the interface is added to the dynamical-inverse-kinematics class.

[kouroshD]

@lrapetti I have open a PR https://github.com/robotology/human-dynamics-estimation/pull/272 on this branch and fix the problem you mentioned in the first comment.

I have verified that Investigate why QP solver in IK does not work for some robots #167 is happening if setCustomConstraintsJointsValues is not called. While addressing point above we should fix also this

[lrapetti]

@kouroshD thanks for the review, I have addressed your suggestions, if there are no further comments I would proceed with merging.

[kouroshD]

Thank you @lrapetti . Looks good to me.

[lrapetti]

Thank you @lrapetti . Looks good to me.

Great, mergin then!