Cholesky failure about g2o HOT 8 CLOSED

Your matrix is not positive semidefinite -- it could be because your graph is not connected. First check if this problem happens with all solvers or only with Cholmod.

from g2o.

I tried the CSparse-Solver, which shows the same behaviour. The used graph should also be connected, but maybe something went wrong during translation from the SLAM-Frontend. Is there a way to get some hints from the matrix in the generated debug.txt?

from g2o.

Can you please check that one of the nodes is fixed, i.e., vertex->setFixed(true) has been called for at least one vertex.

from g2o.

Gnah, I called vertex->fixed() instead of vertex->setFixed(true), a stupid mistake...
Now it completed 5 optimization runs without the error, so this seems to have solved my problem.

from g2o.

Hi
I am facing same problem.
I have checked that (a) Atleast one node is fixed. (b) Graph is connected.
But I am still getting Cholesky failure because matrix is not positive semo definite.
What could be other problem causes?

from g2o.

Be sure that all information matrices are PSD.
You may consider OptimizableGraph::verifyInformationMatrices() for doing
that.

On Tue, Aug 12, 2014 at 11:11 AM, nitin5 [email protected] wrote:

Hi
I am facing same problem.
I have checked that (a) Atleast one node is fixed. (b) Graph is connected.
But I am still getting Cholesky failure because matrix is not positive
semo definite.
What could be other problem causes?

—
Reply to this email directly or view it on GitHub
#35 (comment).

from g2o.

Hi Rainer,
I have checked that all edges' information matrix have PSD property. I have also verified it using OptimizableGraph::verifyInformationMatrices() . I am not able to solve problem till now.

I have debugged it deeper. I will share the source of SIGSEGV . It might be of help in localizing the problem:
In SparseOptimizerIncremental::optimize(), line _underlyingSolver->buildSystem(), control passes to block_solver.hpp::buildSystem(). At line number 532, e->constructQuadraticForm() is called.
From here control passes to Eigen liibrary.
I do not have exposure to Eigen library. In this library, Segmentation fault is reported for functor scalar_sum_op. It is in file Eigen::functors.h line number 26.

I have saved the graph using SparseOptimizerIncremental::save(filename). After removing ONLINE prefix from vertex and edges, I am able to optimise this graph in g2o_viewer, without any error.

Is it okay if i share dataset which i am sunning which others can try to check whether the problem is in my code?
rgds
nitin

from g2o.

For anybody encountering this later, I had a similar error that was due to setting the wrong node to fixed, such that my graph still had gauge freedom.

You can test and fix this with something like:

bool has_gauge_freedom = optimizer->gaugeFreedom();
if (has_gauge_freedom) {
  g2o::OptimizableGraph::Vertex* gauge_node = optimizer->findGauge();
  gauge_node->setFixed(true);
}

from g2o.