🚀 iDynTree 2.0.0 Released

[traversaro]

Hi @robotology/everyone,
iDynTree 2.0.0 was released today, and is now available at https://github.com/robotology/idyntree/releases/tag/v2.0.0 . This is the second major release of iDynTree. As iDynTree adheres to Semantic Versioning, this means that iDynTree 2.0.0 contains API and ABI changes w.r.t. to the iDynTree 1.x releases. iDynTree 2.0.0 will be included in the upcoming iCub-software distro release 2020.11 .

Release Highlights

Some major additions to the library with respect to iDynTree 1.2 are:

Simplify use of KinDynComputations when using non-iDynTree vectors

Since 2015, to avoid any limitation due to the use of a specific Vector or Matrix type, iDynTree has always used its own Vector and Matrix data types in the interface of its classes, for example in iDynTree::KinDynComputations. While this choice had several advantages, it meant that users had to allocate a iDynTree-specific structure whenever they wanted to pass a quantity as input to an iDynTree method, or they wanted to get a result in output. For example, if a user had the the internal joint configuration of their robot in a Eigen::VectorXd, they had to create a separate iDynTree::VectorDynSize, and copy their data in it to call a iDynTree method such as KinDynComputations::setJointPos:

Eigen::VectorXd jointPosInRad;
iDynTree::VectorDynSize jointPosInRad_iDynTree;
iDynTree::KinDynComputations kinDynComp;

// ...

iDynTree::toEigen(jointPosInRad_iDynTree) = jointPosInRad; 
kinDynComp.setJointPos(jointPosInRad_iDynTree);

Since iDynTree 2.0.0, iDynTree supports its non-owning containers for Vector and Matrix, i.e. iDynTree::Span and iDynTree::MatrixView, so that the data can be passed to the methods without unnecessary copies or the creation of intermediate objects, for example to pass Eigen::VectorXd to KinDynComputations::setJointPos :

Eigen::VectorXd jointPosInRad;
iDynTree::KinDynComputations kinDynComp;

// ...

kinDynComp.setJointPos(iDynTree::make_span(jointPosInRad));

This functionality is not limited to Eigen vector, but can be used also with YARP or std vectors. For a more extended example, see
examples/cxx/KinDynComputationsWithEigen/main.cpp.

This feature is available thanks to @GiulioRomualdi and @S-Dafarra .

Python bindings refresh

Python bindings have been improved, for example adding for iDynTree objects the possibility to be constructed from generic iterable objects and NumPy arrays (*.FromPython), and existing iDynTree objects can be converted to NumPy arrays (*.toNumPy).

This improved Python iDynTree bindings are used inside the 1.0 release of gym-ignition, a new framework to create reproducible robotics environments for reinforcement learning research, see #445 .

This feature is available thanks to @diegoferigo .

Computation of the Centroidal Momentum Matrix

The KinDynComputations class has learned how to compute the so-called "Centroidal Momentum Matrix", i.e. the matrix that multiplied on the left for the robot floating base velocity vector, returns the linear and angular momentum expressed in the center of mass of the robot.
This functionality is available in the KinDynComputations::getCentroidalTotalMomentumJacobian method, and is available thanks to the work of @GiulioRomualdi .

Improvements to yarprobostatepublisher

Until iDynTree 2.0.0, the yarprobostatepublisher utility that mimics the capability of the robostatepublisher ROS tool, but just using iDynTree and YARP, published only the transform of each frame of a robot w.r.t. to the base of the robot, producing a "shallow" tree:

While this representation is correct, it does not match the behavior of the robostatepublisher ROS tool, that instead publish the transform of each link frame w.r.t. to the parent link, producing a "deep" tree:

Since iDynTree 2.0, yarprobotstatepublisher has gained a new option, --tree-type with which you can switch between the DEEP and SHALLOW strategy. This feature is available thanks to the work of @elandini84 and @randaz81 .

Detailed ChangeLog

Added

• Added a new CMake option IDYNTREE_COMPILES_TOOLS to disable compilation of iDynTree tools.
• Added a KinDynComputations::getCentroidalTotalMomentumJacobian() method (Implement KinDynComputations::getCentroidalTotalMomentumJacobian() method idyntree#706)
• iDynTree now supports build compiled as a shared library also on Windows.
• When used in Python, new iDynTree objects can be constructed from generic iterable objects and NumPy arrays (*.FromPython),
  and existing objects can be converted to NumPy arrays (*.toNumPy) (Enhance iDynTree from / to Python mapping and add proper NumPy support idyntree#726).
• iDynTree Python bindings can now be installed with pip3 install git+https://github.com/robotology/idyntree.git (Initial support to Python packaging idyntree#733).
• Implement the MatrixView class (Implement the MatrixView class idyntree#734)
• Add the possibility to use MatrixView and Span as input/output objects for KinDynComputations class (Use MatrixView and Span in KinDynComputations class idyntree#736).
• New Python bindings based on pybind11.
  They can be compiled by specifying the CMake option IDYNTREE_USES_PYTHON_PYBIND11. Note that the generated bindings are not
  compatible with the SWIG-generated bindings (e.g. functions have different names). They can be imported as idyntree.pybind Python module.

Fixed

• Fixed bug in yarprobotstatepublisher that caused segmentation fault each time an unknown joint name was read from the input joint states topic (bug-fix: JointIndex validity check idyntree#719)
• Fixed bug in CubicSpline() that causes wrong coefficients calculation when boundary conditions are set (Change the logic used to update the Coefficients in the CubicSpline class idyntree#723)

Changed

• By default iDynTree is compiled as a shared library also on Windows. The BUILD_SHARED_LIBS CMake variable can be used to
  control if iDynTree is built as a shared or a static library.
• The Python method *.fromPyList is replaced by *.FromPython (Enhance iDynTree from / to Python mapping and add proper NumPy support idyntree#726).
• The minimum required CMake version to configure and compile iDynTree is now 3.16 (Bump minimum required version of CMake to 3.16 and update CMake SWIG policies idyntree#732).
• The Python package name of the SWIG bindings changed from iDynTree to idyntree.bindings (Initial support to Python packaging idyntree#733, Update Python SWIG bindings module name idyntree#735). To continue referring to iDynTree classes as iDynTree.<ClassName>, you can change your import iDynTree statements to import idyntree.bindings as iDynTree. Otherwise, you can use import idyntree.bindings to refer them as idyntree.bindings.<ClassName>.
• Improve the use of const keyword in KinDynComputations(Use MatrixView and Span in KinDynComputations class idyntree#736).
• Cleanup size and indices attributes. For consistency with std and Eigen, all sizes and indices have been changed to use std::size_t for unsigned quantities and std::ptrdiff_t for signed quantities. The only exception is the index stored in the triplets of the iDynTree::SparseMatrix data structure, that have been left defined to int for compatibility with Eigen ( Cleanup size and indices attributes, return values and arguments idyntree#767).

Removed

• Remove the CMake option IDYNTREE_USES_KDL and all the classes available when enabling it. They were deprecated in iDynTree 1.0 .
• Remove the semantics related classes. They were deprecated in iDynTree 1.0 .
• Remove unnecessary warning messages from ModelSensorsTransformers.cpp and URDFDocument.cpp (see PR 718)
• Python2 will not be maintained past 2020 and its support has been dropped (Enhance iDynTree from / to Python mapping and add proper NumPy support idyntree#726).
• Remove the need to call iDynTree.init_helpers() and iDynTree.init_numpy_helpers() from Python (Enhance iDynTree from / to Python mapping and add proper NumPy support idyntree#726).
• Remove headers and methods that were deprecated in iDynTree 1.0 (Remove functionalities deprecated in iDynTree 1 idyntree#751).

[traversaro]

A small regression on the second variant of Python bindings (not the one used in the robotology-superbuild's ROBOTOLOGY_USES_PYTHON, to be clear) was found and fixed. This triggered the release of iDynTree 2.0.1 : https://github.com/robotology/idyntree/releases/tag/v2.0.1 .