Issue about persistence of dkpg -i on the TIAGo. After a restart, manually-installed dependency (ros-melodic-pinocchio_2.6.6_amd64 & ros-melodic-hpp-fcl_1.8.1) must be reinstalled.
To do the installation I use the script and sources attached.
scriptInstallDependencies.zip

On pal development computer, to use the library pinocchio, a second eigenpy librairy has been installed:

//the base eigenpy to work with pal's robot environment, cannot remove
ros-melodic-eigenpy/bionic,now 2.5.0-1bionic.20200827.013457 amd64 [installed]

//the eigenpy wanted by pinocchio
robotpkg-py27-eigenpy/bionic,now 2.7.6 amd64 [installed]

When trying to build the package that uses pinocchio, it returns an error relating to the version of eigenpy as it tries to use the wrong eigenpy:

CMake Error at /usr/share/cmake-3.19/Modules/CMakeFindDependencyMacro.cmake:47 (find_package):
  Could not find a configuration file for package "eigenpy" that is
  compatible with requested version "2.6.2".

  The following configuration files were considered but not accepted:

    /opt/ros/melodic/share/eigenpy/cmake/eigenpyConfig.cmake, version: 2.5.0

Call Stack (most recent call first):
  /opt/ros/melodic/share/cmake/pinocchio/pinocchioConfig.cmake:155 (find_dependency)
  CMakeLists.txt:19 (find_package)

In the CMakeLists, the directory of the good eigenpy has already been included.

Find why Pinocchio give slightly wrong values compared to RBDL. Horizontal placements of the arm breaks the gravity compensation (pinocchio::rnea()) with Pinocchio, but not with RBDL.

Pinocchio:

const Eigen::VectorXd &tau = pinocchio::rnea(reduced_model_, data, q_act_, q_zero_, q_zero_);

RBDL:

RigidBodyDynamics::InverseDynamics(rbdl_model_, q_act_, q_zero_, q_zero_, tau_cmd_);

leads: the differences can either be found in the model or in the data (which RBDL does not have)