test-jacobians is not passing #51

nim65s · 2018-10-24T12:34:54Z

Hi,

When trying to run tests, I have this error:

test 4
      Start  4: test-jacobians

4: Test command: /local/users/gsaurel/humanoid-path-planner/hpp-constraints/build/tests/test-jacobians
4: Test timeout computed to be: 9.99988e+06
4: Running 4 test cases...
4: /local/users/gsaurel/humanoid-path-planner/hpp-constraints/tests/test-jacobians.cc(236): error in "jacobian": Jacobian of Configuration seems wrong (central). DOF 21 at row 0: 2.7798 > 0.00012207
4: /local/users/gsaurel/humanoid-path-planner/hpp-constraints/tests/test-jacobians.cc(236): error in "jacobian": Jacobian of Configuration seems wrong (central). DOF 26 at row 0: 3.35197 > 0.00012207
4: /local/users/gsaurel/humanoid-path-planner/hpp-constraints/tests/test-jacobians.cc(236): error in "jacobian": Jacobian of Configuration seems wrong (central). DOF 24 at row 0: 2.94695 > 0.00012207
4: /local/users/gsaurel/humanoid-path-planner/hpp-constraints/tests/test-jacobians.cc(236): error in "jacobian": Jacobian of Configuration seems wrong (central). DOF 24 at row 0: 3.2716 > 0.00012207
4: /local/users/gsaurel/humanoid-path-planner/hpp-constraints/tests/test-jacobians.cc(236): error in "jacobian": Jacobian of Configuration seems wrong (central). DOF 24 at row 0: 3.47788 > 0.00012207
4: 
4: *** 5 failures detected in test suite "hpp_constraints"
 4/10 Test  #4: test-jacobians ...................***Failed    0.18 sec

The text was updated successfully, but these errors were encountered:

jmirabel · 2018-10-24T13:53:08Z

This test passes on my computer... I have no clue about what's going on.

nim65s · 2018-10-24T13:56:15Z

Thanks for your answer, I will try a few things

nim65s · 2018-10-24T15:02:30Z

The bug disappeared when I compiled hpp-pinocchio from the latest release tarball instead of using robotpkg's binary version.

It appeared again when I applied a robotpkg's patch that was meant to allow compilation on 14.04.

I will ask Anthony to remove this patch.

jmirabel · 2018-10-24T15:15:10Z

I understand what happened but I am surprised you did not get an assert. Are you in debug mode ?

To be compliant with old version of Eigen, we should do something like this for all changes in the patch:

if (LgT::NV==Eigen::Dynamic) {
  Jq_.middleRows (row_, lg.nv()).applyOnTheLeft (JqInt)
} else {
  Jq_.middleRows<LgT::NV> (row_).applyOnTheLeft (JqInt)
}

nim65s · 2018-10-24T15:26:23Z

No, I am always in Release mode.
But it is already possible in the CI to launch all tests both in Debug & Release. I should probably activate that on all repo.

I will try to update the patch following your example

jmirabel · 2018-10-24T17:57:11Z

humanoid-path-planner/hpp-pinocchio@749a76a should fix the problem.

jmirabel · 2018-10-24T17:57:43Z

But I cannot easily test it with Eigen 3.2.0.

nim65s · 2018-10-25T07:35:35Z

Thanks a lot, I will try it

nim65s · 2018-10-25T11:16:20Z

This doesn't work, as the operands of the ternary operator must have the same type, and here we have a fixed and a dynamic type:

hpp-pinocchio/src/../src/jdifference-visitor.hh:47:42: error: no match for ternary 'operator?:' (operand types are 'bool', 'Eigen::DenseBase<Eigen::Ref<const Eigen::Matrix<double, -1, 1> > >::SegmentReturnType {aka Eigen::VectorBlock<Eigen::Ref<const Eigen::Matrix<double, -1, 1> >, -1>}', and 'Eigen::DenseBase<Eigen::Ref<const Eigen::Matrix<double, -1, 1> > >::FixedSegmentReturnType<7>::Type {aka Eigen::VectorBlock<Eigen::Ref<const Eigen::Matrix<double, -1, 1> >, 7>}')
_BLOCK_ACCESSOR_EIGEN(q0_,segment,LgT::NQ, iq_, lg.nq()),

I tried a few things without success

nim65s · 2018-10-25T11:17:25Z

for more readability, the types are:

Eigen::DenseBase<Eigen::Ref<const Eigen::Matrix<double, -1, 1> > >::SegmentReturnType                         
Eigen::DenseBase<Eigen::Ref<const Eigen::Matrix<double, -1, 1> > >::FixedSegmentReturnType<7>::Type

jmirabel · 2018-10-25T11:35:39Z

I forgot about that. I fixed it.

nim65s · 2018-10-25T11:38:08Z

Could you push your fix in the topic/eigen32 branch ?

jmirabel · 2018-10-25T11:40:12Z

done.

nim65s · 2018-10-25T11:52:55Z

no more luck:

[…]
hpp-pinocchio/src/../src/dintegrate-visitor.hh:35:11:   required from 'void hpp::pinocchio::liegroupType::dIntegrateVisitor_dq::operator()(const LgT&) [with LgT = se3::SpecialEuclideanOperation<3>]'
[…]
/opt/openrobots/include/pinocchio/spatial/explog.hpp:309:5: error: 'THIS_METHOD_IS_ONLY_FOR_VECTORS_OF_A_SPECIFIC_SIZE' is not a member of 'Eigen::internal::static_assertion<false>'
EIGEN_STATIC_ASSERT_VECTOR_SPECIFIC_SIZE(Vector6Like,6);

jmirabel · 2018-10-25T12:15:27Z

@jcarpent That is a bug in Pinocchio. The problem is that explog does not accept dynamic size vector.

jmirabel · 2018-10-25T13:21:30Z

jmirabel/pinocchio@2d8882c should fix the issue.

nim65s · 2018-10-25T14:34:07Z

So there was 2 typos, fixed in nim65s/pinocchio@400deb5

After that, here is what I get in hpp-constraints in 14.04:

[ 25%] Building CXX object src/CMakeFiles/hpp-constraints.dir/configuration-constraint.cc.o
cd /root/robotpkg/path/hpp-constraints/work.6ab9f1384d29/hpp-constraints-4.2.0/src && /usr/bin/ccache  /usr/bin/g++  -Dhpp_constraints_EXPORTS -pedantic -Wno-long-long -Wall -Wextra -Wcast-align -Wcast-qual -Wformat -Wwrite-strings -Wconversion    -O3 -DNDEBUG -fPIC -I/root/robotpkg/path/hpp-constraints/work.6ab9f1384d29/hpp-constraints-4.2.0 -I/root/robotpkg/path/hpp-constraints/work.6ab9f1384d29/hpp-constraints-4.2.0/include -isystem /usr/include/eigen3 -isystem /opt/openrobots/include -isystem /usr/include/assimp    -DURDFDOM_COLLISION_WITH_GROUP_NAME -DWITH_URDFDOM -DWITH_HPP_FCL -DFCL_USE_ASSIMP_UNIFIED_HEADER_NAMES -DFCL_HAVE_OCTOMAP= -o CMakeFiles/hpp-constraints.dir/configuration-constraint.cc.o -c /root/robotpkg/path/hpp-constraints/work.6ab9f1384d29/hpp-constraints-4.2.0/src/configuration-constraint.cc
/root/robotpkg/path/hpp-constraints/work.6ab9f1384d29/hpp-constraints-4.2.0/src/configuration-constraint.cc: In member function 'virtual void hpp::constraints::ConfigurationConstraint::impl_jacobian(hpp::constraints::matrixOut_t, hpp::constraints::ConfigurationIn_t) const':
/root/robotpkg/path/hpp-constraints/work.6ab9f1384d29/hpp-constraints-4.2.0/src/configuration-constraint.cc:86:93: error: no matching function for call to 'hpp::pinocchio::LiegroupSpace::Jdifference(hpp::constraints::ConfigurationIn_t&, const Eigen::Matrix<double, -1, 1>&, Eigen::Transpose<Eigen::Matrix<double, -1, 1> >, hpp::constraints::matrix_t&)'
goal_.space()->Jdifference<false> (argument, goal_.vector(), diff_.transpose(), unused);
^
/root/robotpkg/path/hpp-constraints/work.6ab9f1384d29/hpp-constraints-4.2.0/src/configuration-constraint.cc:86:93: note: candidate is:
In file included from /opt/openrobots/include/hpp/pinocchio/liegroup-element.hh:20:0,
from /root/robotpkg/path/hpp-constraints/work.6ab9f1384d29/hpp-constraints-4.2.0/include/hpp/constraints/differentiable-function.hh:24,
from /root/robotpkg/path/hpp-constraints/work.6ab9f1384d29/hpp-constraints-4.2.0/include/hpp/constraints/configuration-constraint.hh:25,
from /root/robotpkg/path/hpp-constraints/work.6ab9f1384d29/hpp-constraints-4.2.0/src/configuration-constraint.cc:17:
make[2]: Leaving directory `/root/robotpkg/path/hpp-constraints/work.6ab9f1384d29/hpp-constraints-4.2.0'
/opt/openrobots/include/hpp/pinocchio/liegroup-space.hh:217:12: note: template<bool ApplyOnTheLeft> void hpp::pinocchio::LiegroupSpace::Jdifference(hpp::pinocchio::vectorIn_t, hpp::pinocchio::vectorIn_t, hpp::pinocchio::matrixOut_t, hpp::pinocchio::matrixOut_t) const
void Jdifference (vectorIn_t q0, vectorIn_t q1, matrixOut_t J0, matrixOut_t J1) const;
^
/opt/openrobots/include/hpp/pinocchio/liegroup-space.hh:217:12: note:   template argument deduction/substitution failed:
make[1]: Leaving directory `/root/robotpkg/path/hpp-constraints/work.6ab9f1384d29/hpp-constraints-4.2.0'
/root/robotpkg/path/hpp-constraints/work.6ab9f1384d29/hpp-constraints-4.2.0/src/configuration-constraint.cc:86:93: note:   cannot convert '((Eigen::DenseBase<Eigen::Matrix<double, -1, 1> >*)(&((const hpp::constraints::ConfigurationConstraint*)this)->hpp::constraints::ConfigurationConstraint::diff_))->Eigen::DenseBase<Derived>::transpose<Eigen::Matrix<double, -1, 1> >()' (type 'Eigen::Transpose<Eigen::Matrix<double, -1, 1> >') to type 'hpp::pinocchio::matrixOut_t {aka Eigen::Ref<Eigen::Matrix<double, -1, -1> >}'
goal_.space()->Jdifference<false> (argument, goal_.vector(), diff_.transpose(), unused);
^

jmirabel · 2018-10-25T14:56:46Z

Thanks for the feedback. I pushed the fix on hpp-constraints branch topic/eigen32.

nim65s · 2018-10-25T16:09:14Z

I had this error:

/root/robotpkg/path/hpp-constraints/work.6ab9f1384d29/hpp-constraints-4.2.0/src/configuration-constraint.cc: In member function 'virtual void hpp::constraints::ConfigurationConstraint::impl_jacobian(hpp::constraints::matrixOut_t, hpp::constraints::ConfigurationIn_t) const':
/root/robotpkg/path/hpp-constraints/work.6ab9f1384d29/hpp-constraints-4.2.0/src/configuration-constraint.cc:86:78: error: no matching function for call to 'hpp::pinocchio::LiegroupSpace::Jdifference(hpp::constraints::ConfigurationIn_t&, const Eigen::Matrix<double, -1, 1>&, Eigen::Matrix<double, 1, -1>&, hpp::constraints::matrix_t&)'
goal_.space()->Jdifference<false> (argument, goal_.vector(), J_, unused);
^
/root/robotpkg/path/hpp-constraints/work.6ab9f1384d29/hpp-constraints-4.2.0/src/configuration-constraint.cc:86:78: note: candidate is:
In file included from /opt/openrobots/include/hpp/pinocchio/liegroup-element.hh:20:0,
from /root/robotpkg/path/hpp-constraints/work.6ab9f1384d29/hpp-constraints-4.2.0/include/hpp/constraints/differentiable-function.hh:24,
from /root/robotpkg/path/hpp-constraints/work.6ab9f1384d29/hpp-constraints-4.2.0/include/hpp/constraints/configuration-constraint.hh:25,
from /root/robotpkg/path/hpp-constraints/work.6ab9f1384d29/hpp-constraints-4.2.0/src/configuration-constraint.cc:17:
/opt/openrobots/include/hpp/pinocchio/liegroup-space.hh:217:12: note: template<bool ApplyOnTheLeft> void hpp::pinocchio::LiegroupSpace::Jdifference(hpp::pinocchio::vectorIn_t, hpp::pinocchio::vectorIn_t, hpp::pinocchio::matrixOut_t, hpp::pinocchio::matrixOut_t) const
void Jdifference (vectorIn_t q0, vectorIn_t q1, matrixOut_t J0, matrixOut_t J1) const;
^
/opt/openrobots/include/hpp/pinocchio/liegroup-space.hh:217:12: note:   template argument deduction/substitution failed:
/root/robotpkg/path/hpp-constraints/work.6ab9f1384d29/hpp-constraints-4.2.0/src/configuration-constraint.cc:86:78: note:   cannot convert '((const hpp::constraints::ConfigurationConstraint*)this)->hpp::constraints::ConfigurationConstraint::J_' (type 'Eigen::Matrix<double, 1, -1>') to type 'hpp::pinocchio::matrixOut_t {aka Eigen::Ref<Eigen::Matrix<double, -1, -1> >}'
goal_.space()->Jdifference<false> (argument, goal_.vector(), J_, unused);
^

So I changed your mutable Eigen::Matrix<value_type, 1, Eigen::Dynamic> J_; by a mutable Eigen::Matrix<value_type, Eigen::Dynamic, Eigen::Dynamic> J_;

And now I have a bunch of:

hpp-constraints/src/static-stability.cc:173:35: error: 'class Eigen::JacobiSVD<Eigen::Matrix<double, -1, -1>, 2>' has no member named 'rank'

It turns out that the rank() method has not been added in Eigen before 3.2.2

jmirabel · 2018-10-25T17:32:34Z

It's gonna be tedious to be compatible with Eigen 3.2.0. In 14.04 previously, we installed a more recent version of Eigen.

Let's not put HPP in robotpkg for Ubuntu 14.04 (or at least keep this for later). I think @florent-lamiraux will agree.

nim65s · 2018-10-25T17:49:09Z

I agree.

I will just cc @pFernbach here, because I think that he needs to get a few things to work with the current HPP in 14.04 before being able to upgrade to 16.04.
Anyway, in his case, installing a more recent version of Eigen in 14.04 might help.

Still, should we open PRs for:

humanoid-path-planner/hpp-pinocchio@devel...topic/eigen32
stack-of-tasks/pinocchio@devel...nim65s:matrix_size (after a merge of Split Jdifference in dDifference_dq<0> and dDifference_dq<1> stack-of-tasks/pinocchio#561)
f99ef04 with Eigen::Dynamic (this one I am less sure…)

pFernbach · 2018-10-25T18:13:15Z

We changed our mind and I will stop trying to use the current devel/master on 14.04.
We are also going to switch rbprm to 16.

jmirabel · 2018-10-25T20:20:56Z

For the one in pinocchio, yes, we can open a PR.
I'll deal with the two others.

nim65s · 2018-10-25T20:26:02Z

perfect, thanks :)

jmirabel · 2019-01-25T09:05:08Z

Is this still ongoing ?

jmirabel closed this as completed Feb 8, 2019

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

test-jacobians is not passing #51

test-jacobians is not passing #51

nim65s commented Oct 24, 2018

jmirabel commented Oct 24, 2018

nim65s commented Oct 24, 2018

nim65s commented Oct 24, 2018 •

edited

jmirabel commented Oct 24, 2018

nim65s commented Oct 24, 2018

jmirabel commented Oct 24, 2018

jmirabel commented Oct 24, 2018

nim65s commented Oct 25, 2018

nim65s commented Oct 25, 2018

nim65s commented Oct 25, 2018 •

edited

jmirabel commented Oct 25, 2018

nim65s commented Oct 25, 2018

jmirabel commented Oct 25, 2018

nim65s commented Oct 25, 2018

jmirabel commented Oct 25, 2018

jmirabel commented Oct 25, 2018

nim65s commented Oct 25, 2018 •

edited

jmirabel commented Oct 25, 2018 •

edited

nim65s commented Oct 25, 2018 •

edited

jmirabel commented Oct 25, 2018

nim65s commented Oct 25, 2018 •

edited

pFernbach commented Oct 25, 2018

jmirabel commented Oct 25, 2018

nim65s commented Oct 25, 2018

jmirabel commented Jan 25, 2019

test-jacobians is not passing #51

test-jacobians is not passing #51

Comments

nim65s commented Oct 24, 2018

jmirabel commented Oct 24, 2018

nim65s commented Oct 24, 2018

nim65s commented Oct 24, 2018 • edited

jmirabel commented Oct 24, 2018

nim65s commented Oct 24, 2018

jmirabel commented Oct 24, 2018

jmirabel commented Oct 24, 2018

nim65s commented Oct 25, 2018

nim65s commented Oct 25, 2018

nim65s commented Oct 25, 2018 • edited

jmirabel commented Oct 25, 2018

nim65s commented Oct 25, 2018

jmirabel commented Oct 25, 2018

nim65s commented Oct 25, 2018

jmirabel commented Oct 25, 2018

jmirabel commented Oct 25, 2018

nim65s commented Oct 25, 2018 • edited

jmirabel commented Oct 25, 2018 • edited

nim65s commented Oct 25, 2018 • edited

jmirabel commented Oct 25, 2018

nim65s commented Oct 25, 2018 • edited

pFernbach commented Oct 25, 2018

jmirabel commented Oct 25, 2018

nim65s commented Oct 25, 2018

jmirabel commented Jan 25, 2019

nim65s commented Oct 24, 2018 •

edited

nim65s commented Oct 25, 2018 •

edited

nim65s commented Oct 25, 2018 •

edited

jmirabel commented Oct 25, 2018 •

edited

nim65s commented Oct 25, 2018 •

edited

nim65s commented Oct 25, 2018 •

edited