Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Gpm null space dimension print #126

Closed
wants to merge 5 commits into from
Closed

Gpm null space dimension print #126

Show file tree
Hide file tree
Changes from 2 commits
Commits
Show all changes
5 commits
Select commit Hold shift + click to select a range
58be1d3
Null space projection matrix rank is analised to avoid unnecessary ca…
Jan 12, 2017
dbe568e
adding warining messages in projection matrix of the null space all n…
Jan 13, 2017
12c5bba
if condition splited as in pull request comment
Jan 13, 2017
95994a0
code optimisation to compare the projector rank with zero matrix
Jan 16, 2017
6bf88eb
bug fix
Jan 16, 2017
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
30 changes: 20 additions & 10 deletions cob_twist_controller/src/constraint_solvers/solvers/gradient_projection_method_solver.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -52,19 +52,29 @@ Eigen::MatrixXd GradientProjectionMethodSolver::solve(const Vector6d_t& in_cart_

Eigen::MatrixXd homogeneous_solution = Eigen::MatrixXd::Zero(particular_solution.rows(), particular_solution.cols());
KDL::JntArrayVel predict_jnts_vel(joint_states.current_q_.rows());
Eigen::FullPivLU<Eigen::MatrixXd> lu_decomp(projector);
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Performance question:
I think it is quite expensive to perform a FullPivLU to determine rank
Could we not just check whether the determinant of a PartialPivLU is 0.0?

Or even easier:
compare projector with a Zero-Matrix using isApprox

Eigen::MatrixXd qdots_out;

for (std::set<ConstraintBase_t>::iterator it = this->constraints_.begin(); it != this->constraints_.end(); ++it)
if(lu_decomp.rank() != 0)
{
ROS_DEBUG_STREAM("task id: " << (*it)->getTaskId());
(*it)->update(joint_states, predict_jnts_vel, this->jacobian_data_);
Eigen::VectorXd q_dot_0 = (*it)->getPartialValues();
Eigen::MatrixXd tmp_projection = projector * q_dot_0;
double activation_gain = (*it)->getActivationGain(); // contribution of the homo. solution to the part. solution
double constraint_k_H = (*it)->getSelfMotionMagnitude(particular_solution, tmp_projection); // gain of homogenous solution (if active)
homogeneous_solution += (constraint_k_H * activation_gain * tmp_projection);
}

Eigen::MatrixXd qdots_out = particular_solution + this->params_.k_H * homogeneous_solution; // weighting with k_H is done in loop
for (std::set<ConstraintBase_t>::iterator it = this->constraints_.begin(); it != this->constraints_.end(); ++it)
{
ROS_DEBUG_STREAM("task id: " << (*it)->getTaskId());
(*it)->update(joint_states, predict_jnts_vel, this->jacobian_data_);
Eigen::VectorXd q_dot_0 = (*it)->getPartialValues();
Eigen::MatrixXd tmp_projection = projector * q_dot_0;
double activation_gain = (*it)->getActivationGain(); // contribution of the homo. solution to the part. solution
double constraint_k_H = (*it)->getSelfMotionMagnitude(particular_solution, tmp_projection); // gain of homogenous solution (if active)
homogeneous_solution += (constraint_k_H * activation_gain * tmp_projection);
}

qdots_out = particular_solution + this->params_.k_H * homogeneous_solution; // weighting with k_H is done in loop
}
else{
qdots_out = particular_solution;
ROS_WARN("Null space projection matrix is null. The constraint may not be satisfied");
}

// //DEBUG: for verification of nullspace projection
// std::stringstream ss_part;
Expand Down
17 changes: 14 additions & 3 deletions cob_twist_controller/src/constraint_solvers/solvers/stack_of_tasks_solver.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -106,12 +106,23 @@ Eigen::MatrixXd StackOfTasksSolver::solve(const Vector6d_t& in_cart_velocities,
Eigen::MatrixXd J_temp = J_task * projector_i;
Eigen::VectorXd v_task = it->task_;
Eigen::MatrixXd J_temp_inv = pinv_calc_.calculate(J_temp);
Eigen::FullPivLU<Eigen::MatrixXd> lu_decomp(J_temp_inv);
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Why do you evaluate J_temp_inv and not projector_i?

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Because J_temp_inv is going to project the ith task velocities. If it has rank zero is not adding any velocity for that task.

if(lu_decomp.rank()== 0){
qdots_out.col(0) = q_i;
ROS_WARN("Null space projection matrix is null. It couldn't satisfy all constraints");
return qdots_out;
}
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I think we cannot quit here as projector_i = projector_i - J_temp_inv * J_temp; so even if some projector_i are Zero, the overall projector_i not necessarily is Zero as it is a sum not a multiplication.

Just a warning that the projector of the current task is Zero but continue computation

q_i = q_i + J_temp_inv * (v_task - J_task * q_i);
projector_i = projector_i - J_temp_inv * J_temp;
}

qdots_out.col(0) = q_i + projector_i * sum_of_gradient;
return qdots_out;
Eigen::FullPivLU<Eigen::MatrixXd> lu_decomp(projector_i);
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

same performance comment as above

if(lu_decomp.rank()== 0){
qdots_out.col(0) = q_i;
ROS_WARN("Null space projection matrix is null. It couldn't satisfy the global weighting for all constraints.");
return qdots_out;
}
qdots_out.col(0) = q_i + projector_i * sum_of_gradient; //WHY???
return qdots_out;
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Why //WHY???

identation

}


Expand Down
4 changes: 3 additions & 1 deletion cob_twist_controller/src/constraint_solvers/solvers/task_priority_solver.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -64,8 +64,9 @@ Eigen::MatrixXd TaskPrioritySolver::solve(const Vector6d_t& in_cart_velocities,
predict_jnts_vel.q(i) = particular_solution(i, 0) * cycle + joint_states.current_q_(i);
predict_jnts_vel.qdot(i) = particular_solution(i, 0);
}
Eigen::FullPivLU<Eigen::MatrixXd> lu_decomp(projector);
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

same performance question


if (this->constraints_.size() > 0)
if ((this->constraints_.size() > 0) && (lu_decomp.rank() != 0))
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Can you put the check into a new if?
Like this:

if (this->constraints_.size() > 0)
{
    if(lu_decomp.rank() != 0)
    {

because than it's more clear was has happened, i.e. whether the first or the second condition is false

{
for (std::set<ConstraintBase_t>::iterator it = this->constraints_.begin(); it != this->constraints_.end(); ++it)
{
Expand Down Expand Up @@ -94,6 +95,7 @@ Eigen::MatrixXd TaskPrioritySolver::solve(const Vector6d_t& in_cart_velocities,
{
qdots_out = particular_solution;
ROS_ERROR_STREAM("Should not occur solution: " << std::endl << qdots_out);
ROS_WARN("Null space projection matrix is null. The constraint may not be satisfied");
}

// Eigen::MatrixXd qdots_out = particular_solution + homogeneousSolution; // weighting with k_H is done in loop
Expand Down