TSID - Task Space Inverse Dynamics

TSID is a C++ library for optimization-based inverse-dynamics control based on the rigid multi-body dynamics library Pinocchio.

Documentation

• Take a look at the project wiki for an overview of the design of the library.
• In the exercises folder you can find several examples of how to use TSID in Python with robot manipulators, humanoids, or quadrupeds.
• On the website of Andrea Del Prete you can find slides and video lessons on TSID.
• Memmo 2020 summer school

Installation with Conda

If you want to directly dive into TSID in Python, only one single line is sufficient (assuming you have Conda installed):

conda install tsid -c conda-forge

Installation from Debian/Ubuntu packages, with robotpkg

If you have never added robotpkg’s software repository you can do it with the following commands:

sudo tee /etc/apt/sources.list.d/robotpkg.list <<EOF
deb [arch=amd64] http://robotpkg.openrobots.org/packages/debian/pub $(lsb_release -sc) robotpkg
EOF

curl http://robotpkg.openrobots.org/packages/debian/robotpkg.key | sudo apt-key add -
sudo apt update

You can install TSID and its python bindings (replace * with you Python version) with:

sudo apt install robotpkg-py3*-tsid

Installation from sources

First you need to install the following dependencies:

• boost (unit_test_framework)
• eigen3
• pinocchio
• eiquadprog
• example-robot-data (only for running the examples)

To install eigen3 on Ubuntu you can use apt-get: sudo apt-get install libeigen3-dev

To install pinocchio follow the instruction on its website.

To compile TSID:

cd $DEVEL/openrobots/src/
git clone --recursive git@github.com:stack-of-tasks/tsid.git
cd tsid
mkdir _build-RELEASE
cd _build-RELEASE
cmake .. -DCMAKE_BUILD_TYPE=RELEASE -DCMAKE_INSTALL_PREFIX=$DEVEL/openrobots
make install

Python Bindings

To use this library in python, we offer python bindings based on Boost.Python and EigenPy.

To install EigenPy you can compile the source code:

git clone https://github.com/stack-of-tasks/eigenpy

or, on Ubuntu, you can use apt-get:

sudo apt-get install robotpkg-py3*-eigenpy

For testing the python bindings, you can run the unit test scripts in the script folder, for instance:

ipython script/test_formulation.py

To run the demo using gepetto-viewer:

ipython demo/demo_romeo.py

Credits

This package is authored by:

• Andrea Del Prete (University of Trento)
• Justin Carpentier (INRIA)

It includes key contributions from:

• Julian Viereck (Max Planck Institute, New York University)
• Sanghyun Kim (Seoul National University)
• Eloise Dalin (LORIA, INRIA Lorraine)
• Noelie Ramuzat (LAAS, CNRS)
• Pierre Fernbach (LAAS, CNRS)
• Aurelie Bonnefoy (LAAS, CNRS)
• Etienne Arlaud (INRIA)
• Fabian Schramm (INRIA)

And is maintained by:

• Guilhem Saurel (LAAS-CNRS)

Citing

If you are (or not) happy with TSID and want to cite it, please use the following citation:

@inproceedings {adelprete:jnrh:2016,
    title = {Implementing Torque Control with High-Ratio Gear Boxes and without Joint-Torque Sensors},
    booktitle = {Int. Journal of Humanoid Robotics},
    year = {2016},
    pages = {1550044},
    url = {https://hal.archives-ouvertes.fr/hal-01136936/document},
    author = {Andrea Del Prete, Nicolas Mansard, Oscar E Ramos, Olivier Stasse, Francesco Nori}
}

Changelog

All notable changes to this project will be documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

Unreleased

1.7.1 - 2024-08-26

• Fix a typo in ex_4_walking
• check solver status, set eps_abs to 1e-6, fix seed
• CMake: require >= 3.10
• add changelog
• setup ruff & fix isort pre-commit config
• update ROS CI

1.7.0 - 2023-05-13

• expose SE3ToVector and vectorToSE3
• remove warnings
• Enhance Python target packaging
• Add CI with GitHub Action for conda
• Add support for proxqp and osqp solver
• Clean and update contributors list
• pre-commit autoupdate
• Expose TaskJointPosVelAccBounds
• Expose AddMotionTask for TaskJointPosVelAccBounds
• Fix qpmad
• Add clang-format Google style
• Add Measured force as an external force task with moving objects
• update CMake: fetch submodule, set default build type, fix RPATH
• fix for eigenpy v3
• np.matrix → np.array
• tooling: setup black isort & toml-sort

1.6.3 - 2022-11-02

• Require C++17
• fix tests in 18.04
• update pinocchio use

1.6.2 - 2022-09-05

• update python tests for np array
• add task actuation equality
• use generated headers
• Choose floating-base (or not) when creating a robot wrapper
• Add optional support of qpmad solver
• update packaging for eigenpy 2.7.12

1.6.1 - 2021-10-19

This new release is mostly a maintenance release, including fixes for the recent version of Eigen 3.4.

1.6.0 - 2021-03-18

• Add bindings for method Contact6d::getMotionTask
• Add Center of Pressure Taks
• fix warnings

1.5.0 - 2021-03-03

• [py] Add config variable to specify whether end-effector task should be formulated in local frame or world frame.
• [C++] Standardize names of methods to get and set mask in motion tasks.
• [py] Add accessors masks in TaskCOM and TaskSE3Equality.
• bugfixes

1.4.2 - 2020-11-26

• Add a mask to task-com-equality
• Fix bug in Contact6d::setRegularizationTaskWeightVector
• CMake: fix configure without tests

1.4.1 - 2020-09-25

• fix memory leaks thanks to shared_ptr
• fix warnings
• fix package.xml for ROS

1.4.0 - 2020-09-09

• add setGravity
• stop using StdVec in python to make code more user friendly
• add 6d contact with motion constraint at priority level 1 in python
• use example-robot-data in notebooks

1.3.1 - 2020-06-05

• fix license
• fix generated tsid.pc

1.3.0 - 2020-05-26

• reactive test
• add package.xml
• add CheatSheet
• updates for numpy.array & python 3
• updateRigidContactWeights: fix and add to python API
• use eiquadprog
• fix compatibility with pinocchio v2.4.5

1.2.3 - 2020-03-30

• renamed tests dir
• fix python tests
• CMake: export project and use exports from dependencies
• CMake: keep minimal required instructions

1.2.2 - 2020-03-01

• add angular momentum equality task
• update to pinocchio changes
• Python 3 compatibility
• add some documentation
• fix python issue

1.2.1 - 2019-09-19

• fix compatibility with recent pinocchio versions

1.2.0 - 2019-03-06

• Pinocchio v2, fix #31
• Fix demo_romeo for pinocchio v2
• Pull request for use TSID in python.
• Add missing includes, fix #18

1.1.0 - 2018-10-10

This release updates to non backward-compatibles changes in pinocchio v1.3.0

1.0.2 - 2018-06-12

This release is mostly a maintenance release. It fixes some bug with respect to Pinocchio. It also fixes some issues with respect to the packaging.

1.0.1 - 2018-01-12

• [joint-posture-task] Fix bug in computation of task matrix from mask
• [task-se3-equality] Add method to get frame-id
• [Robot] Add missing evaluation of the center of mass acceleration provided zero joint acceleration
• [Math] Fix bug related to Eigen undefined function set_is_malloc_allowed
• [CMake] Correct minimal version of Eigen3
• [inv-dyn-form-acc-force] Fix potential bug in removal of contact constraint
• [task-se3-equality] Fix small bug in computaiton of acceleration (just used for debugging)
• [contact-6d] Add methods to set reference force and weight vector
• [inv-dyn-form-acc-force] Fix bug: weight of force regularization task was not updated
• [robot-wrapper] Fix bug in mass matrix: copy upper triangular part to lower triangular part (before this it was set to zero)
• [formulations] Remove debug prints
• [formulations] Fix bug in update of task weights
• [robot-wrapper] BUG FIX: compute center of mass acceleration in computeAllTerms (before it was not computed so we were introducing random number in the CoM task)
• [tsid-formulations] Add method to change the weight of a task.
• [eigquadprog-fast] In DEBUG, in case a constraint is not verified, set the status flag to ERROR (even if the solver said the problem has been solved)
• [task-com-equality] Fix little bugs in methods to get pos/vel references and errors, and in method to get desired acceleration
• [math-utils] Add function to check if matrix/vector contains NaN
• [math-utils] Pass JacobiSVD by reference in function solveWithDampingFromSvd
• [math-utils] Add function to solve linear system of equations from svd decomposition

1.0.0 - 2017-06-16

This is the first release of TSID. This release includes minimal features for the torque control of humanoid robots such as HRP-2.