QP_IK_solver

This repository includes the packages and instructions to solve the centralized inverse kinematic problem, formulated as a quadratic program (QP) and solved in real-time. You can find the paper here:

and the corresponding video here:

Dependencies

Eigen http://eigen.tuxfamily.org/index.php?title=Main_Page

Svm_grad https://github.com/nbfigueroa/SVMGrad.git

Sg_differentiation https://github.com/epfl-lasa/sg_differentiation.git

Nlopt http://ab-initio.mit.edu/wiki/index.php/NLopt

CVXGEN http://cvxgen.com/docs/index.html (Due to the license, I needed to remove CVxgen, please follow the steps in the mentioned webpage to get the CVxgen files.)

VERY Very important NOTE: This package will not work without CVXGEN! You must the files and put them in a folder named cvxgen.

Features:

• Handling the kinematic constraints of the robots.
• Inequity constraint (We used it for considering the self-collision avoidance (SCA) constraints).
• This package provides three options for solving the inverse kinematic:
  • LVI-based primal-dual Dynamical system solution: http://ieeexplore.ieee.org/abstract/document/1470152/
  • Nlopt
  • CVXgen

How to run

Initialization:

1-Initialize the IK solver: 1-1 If You don't want to use the SCA constraint

Initialize(Number of robots ,Time sample ,Solver_type={Numerical=0,Dynamical},Solver_level={Velocity_level=0,Acceleration_level},Super_constraint={True, False})

1-2 If You want to use the SCA constraint

Initialize(Number of robots ,Time sample ,Solver_type={Numerical=0,Dynamical},Solver_level={Velocity_level=0,Acceleration_level},Super_constraint={True, False},svm_filename)

The Dynamical solver uses the LVI-based primal-dual Dynamical system solver and the default Numerical solver is CVXGEN. If you want to change it to Nlopt you need to modify qp_ik_solver.h. svm_filename is a path to the learned SVR collision configurations. 2- Initialize each robot:

Initialize_robot(index of the robot,Number of links,Number of end-effector constraints,W,Upper bound of joints' position, Lower bound of joints' position,Upper bound of joints' velocity,Lower bound of joints' velocity, Upper bound of joints' acceleration, Lower bound of joints' acceleration)

3- Finalize initialization

Finalize_Initialization()

In the loop:

4- In the loop:

set_jacobian(index of the robot,Jacobian of the end-effector)
set_jacobian_links(index of the robot,Jacobian and Cartesian position of all the links)
set_state(index of the rooot,current joint position,current joint velocity);
set_desired(index of the robot,Desired end-effector state);
Solve();
get_state(index of the robot, desired joint velocity);

Copyright

Please cite these papers if you are using this toolbox: @article{mirrazavi2018unified, title={A unified framework for coordinated multi-arm motion planning}, author={Mirrazavi Salehian, Seyed Sina and Figueroa, Nadia and Billard, Aude}, journal={The International Journal of Robotics Research}, pages={0278364918765952}, publisher={SAGE Publications Sage UK: London, England} }

For more information contact Sina Mirrazavi.