Skip to content

Latest commit

 

History

History
186 lines (152 loc) · 9.33 KB

README.md

File metadata and controls

186 lines (152 loc) · 9.33 KB

robotoc - efficient ROBOT Optimal Control solvers

build codecov

     

     

Features for efficient optimal control of robotic systems

  • Direct multiple-shooting method based on the lifted contact dynamics / inverse dynamics.
  • Riccati recursion algorithm for switching time optimization (STO) problems and efficient pure-state equality constraint handling.
  • Primal-dual interior point method for inequality constraints.
  • Very fast computation of rigid body dynamics and its sensitivities thanks to Pinocchio.

Requirements

  • Ubuntu 22.04, 20.04, and 18.04 (possibly Mac OSX, but not well tested)
  • gcc (at least C++11 is required), CMake (at least version 3.11)
  • Eigen3, Pinocchio ,
  • Python3, NumPy (for Python binding)
  • gepetto-viewer-corba and/or meshcat-python (optional to visualize the solution trajectory in Python)
  • PyBullet (optional to simulate MPC examples)

Installation

  1. Install the latest stable version of Eigen3 by
sudo apt install libeigen3-dev
  1. Install the latest stable version of Pinocchio by following the instruction
  2. Clone this repository and change the directory as
git clone https://github.com/mayataka/robotoc
cd robotoc 
  1. Build and install robotoc as
mkdir build && cd build
cmake .. -DCMAKE_BUILD_TYPE=Release 
make install -j$(nproc)

NOTE: if you want to maximize the performance, use CMake option

cmake .. -DCMAKE_BUILD_TYPE=Release -DOPTIMIZE_FOR_NATIVE=ON
  1. If you want to visualize the solution trajectory with Python, you have to install gepetto-viewer-corba by
sudo apt update && sudo apt install robotpkg-py38-qt5-gepetto-viewer-corba -y

and/or meshcat-python by

pip install meshcat
  1. If you do not want to install the Python bindings, change the CMake configuration as
cmake .. -DBUILD_PYTHON_INTERFACE=OFF
  1. In OSX, explicitly set g++ as the complier. First, find the path of g++ as
ls -l /usr/local/bin | grep g++

Then set the full path in the cmake as

cmake .. -DCMAKE_CXX_COMPILER=FULL_PATH_TO_GPLUSPLUS

Usage

C++:

You can link your executables to robotoc by writing CMakeLists.txt, e.g., as

find_package(robotoc REQUIRED)

add_executable(
    YOUR_EXECTABLE
    YOUR_EXECTABLE.cpp
)
target_link_libraries(
    YOUR_EXECTABLE
    PRIVATE
    robotoc::robotoc
)

Python:

Suppose that the Python version is 3.8. The Python bindings will then be installed at ROBOTOC_INSTALL_DIR/lib/python3.8/site-packages where ROBOTOC_INSTALL_DIR is the install directory of robotoc configured in CMake (e.g., by -DCMAKE_INSTALL_PREFIX). To use the installed Python library, it is convenient to set the environment variable as

export PYTHONPATH=ROBOTOC_INSTALL_DIR/lib/python3.8/site-packages:$PYTHONPATH 

e.g., in ~/.bashrc. Note that if you use another Python version than python3.8, please adapt it.

Solvers

The following three solvers are provided:

  • OCPSolver : Solves the OCP for rigid-body systems (possibly with contacts) by using Riccati recursion. Can optimize the switching times and the trajectories simultaneously.
  • UnconstrOCPSolver : Solves the OCP for "unconstrained" rigid-body systems by using Riccati recursion.
  • UnconstrParNMPCSolver : Solves the OCP for "unconstrained" rigid-body systems by using ParNMPC algorithm.

where "unconstrained" rigid-body systems are systems without any contacts or a floating-base.

Examples

Examples of these solvers are found in examples directory. Further explanations are found at https://mayataka.github.io/robotoc/page_examples.html.

Switching time optimization (STO) examples

  • OCPSolver can solve the switching time optimization (STO) problem, which optimizes the trajectory and the contact timings simultaneously.
  • The following videos display the solution trajectory of the STO problems (anymal/python/jumping_sto.py and icub/python/jumping_sto.py).

 

Whole-body MPC examples

  • The following example implementations of whole-body MPC are provided:
    • MPCCrawl : MPC with OCPSolver for the crawl gait of quadrupedal robots.
    • MPCTrot : MPC with OCPSolver for the trot gait of quadrupedal robots.
    • MPCPace : MPC with OCPSolver for the pace gait of quadrupedal robots.
    • MPCFlyingTrot : MPC with OCPSolver for the flying trot gait of quadrupedal robots.
    • MPCJump : MPC with OCPSolver for the jump motion of quadrupedal or bipedal robots.
    • MPCBipedWalk : MPC with OCPSolver for the walking motion of bipedal robots.
  • You can find the simulations of these MPC at a1/mpc, anymal/mpc, and icub/mpc.
  • You need PyBullet for the MPC simulations. (easy to install, e.g., by pip install pybullet)
  • The following videos display the simulation results of quadrupedal walking on rough terrain (a1/mpc/trot_terrain.py) and bipedal walking (icub/mpc/walk.py).

 

Documentation

More detailed documentation is available at https://mayataka.github.io/robotoc/.

Tutorial

Our tutorials on robotoc is available at robotoc_tutorial.

Citing robotoc

  • Citing the STO algorithm of OCPSolver:
@misc{katayama2021sto,
  title={Structure-exploiting {N}ewton-type method for optimal control of switched systems}, 
  author={Sotaro Katayama and Toshiyuki Ohtsuka},
  url={arXiv:2112.07232},
  eprint={2112.07232},
  archivePrefix={arXiv}
  year={2021}}
  • Citing OCPSolver without the switching time optimization (STO):
@inproceedings{katayama2022lifted,
  title={Lifted contact dynamics for efficient optimal control of rigid body systems with contacts}, 
  author={Sotaro Katayama and Toshiyuki Ohtsuka},
  booktitle={{2022 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS 2022) (to appear)}},
  year={2022}}
@inproceedings{katayama2021idocp,
  title={Efficient solution method based on inverse dynamics for optimal control problems of rigid body systems},
  author={Sotaro Katayama and Toshiyuki Ohtsuka},
  booktitle={{2021 IEEE International Conference on Robotics and Automation (ICRA)}},
  pages={2070--2076},
  year={2021}}

Related publications

  • S. Katayama and T. Ohtsuka, "Whole-body model predictive control with rigid contacts via online switching time optimization," 2022 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS 2022) (to appear), https://arxiv.org/abs/2203.00997, 2022
  • S. Katayama and T. Ohtsuka, "Structure-exploiting Newton-type method for optimal control of switched systems," https://arxiv.org/abs/2112.07232, 2021
  • S. Katayama and T. Ohtsuka, Lifted contact dynamics for efficient optimal control of rigid body systems with contacts, 2022 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) (to appear), https://arxiv.org/abs/2108.01781, 2022
  • S. Katayama and T. Ohtsuka, Efficient Riccati recursion for optimal control problems with pure-state equality constraints, 2022 American Control Conference (ACC), pp. 3579-3586, 2022
  • S. Katayama and T. Ohtsuka, Efficient solution method based on inverse dynamics for optimal control problems of rigid body systems, 2021 IEEE International Conference on Robotics and Automation (ICRA), pp.2070-2076, 2021
  • H. Deng and T. Ohtsuka, A parallel Newton-type method for nonlinear model predictive control, Automatica, Vol. 109, pp. 108560, 2019