Error in evaluating the ResidualBlock

[taiping]

Hi,

I use linear solver SPARSE_NORMAL_CHOLESKY to solve graph pose slam problem. It works fine when residuals are small. However, when loop closure occurs, I get errors.

From A to A`, it works fine. Since A and A` is the same location, the loop closure occurs and I get errors.

For the similar error in https://github.com/googlecartographer/cartographer/issues/233#issuecomment-294993684,

the solution is to use EigenSparse instead of SuiteSparse for a bug in SuiteSparse.

I tried linear solver DENSE_NORMAL_CHOLESKY and DENSE_QR based on EigenSparse. It is very slow, and I also get same errors.

I hope I make it clearly.

The error I've been getting is shown below:

W0603 10:40:14.056905 8994 residual_block.cc:131]

Error in evaluating the ResidualBlock.

There are two possible reasons. Either the CostFunction did not evaluate and fill all

residual and jacobians that were requested or there was a non-finite value (nan/infinite)

generated during the or jacobian computation.

Residual Block size: 7 parameter blocks x 3 residuals

For each parameter block, the value of the parameters are printed in the first column

and the value of the jacobian under the corresponding residual. If a ParameterBlock was

held constant then the corresponding jacobian is printed as 'Not Computed'. If an entry

of the Jacobian/residual array was requested but was not written to by user code, it is

indicated by 'Uninitialized'. This is an error. Residuals or Jacobian values evaluating

to Inf or NaN is also an error.

Residuals: 0 0 0

Parameter Block 0, size: 1

-48.3575 | -nan -nan -nan

Parameter Block 1, size: 1

-428.744 | -nan -nan -nan

Parameter Block 2, size: 1

1.11116 | -nan -nan -nan

Parameter Block 3, size: 1

-28.9954 | -nan -nan -nan

Parameter Block 4, size: 1

-250.725 | -nan -nan -nan

Parameter Block 5, size: 1

1.27572 | -nan -nan -nan

Parameter Block 6, size: 1

-65.0826 | -nan -nan -nan

E0603 10:40:14.057034 8994 trust_region_minimizer.cc:113] Terminating: Residual and Jacobian evaluation failed.

Best,

Taiping

[Sameer Agarwal]

Taiping,

This indicates something is wrong with your computation of the Jacobian. It has nothing to do with the sparse linear algebra library. You should check the cost function and how it is computing the Jacobian.

Sameer

--
You received this message because you are subscribed to the Google Groups "Ceres Solver" group.
To unsubscribe from this group and stop receiving emails from it, send an email to ceres-solver...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/ceres-solver/e8860fd9-2bcd-4e44-830b-fe741296d129%40googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[Eugen Funk]

Dear Sameer,

I am having similar issue when using AutoDiffCostFunction.

The problem mainly arises when I set the initial values to the exact solution. Thus, the residuum is 0 and the jacobians are alle nan.

Is there a way to debug the reason for this? Is the trust region too small?

Best

Eugen

[Sameer Agarwal]

Eugen,
Have a look at the stderr, there is detailed logging information that is recorded when a Jacobian computation produces nans. 

Sameer

To view this discussion on the web visit https://groups.google.com/d/msgid/ceres-solver/10db297f-462c-46fb-bdbe-264737a3456a%40googlegroups.com.

[William Rucklidge]

One thing to look at: does your residual include a square root (or other fractional power)? If it does, you may run into trouble when the residual is zero, or very close to it, as the derivative of the square root function is infinite at zero, which can quite easily lead to NaNs in the Jacobian. This can be a problem for analytical derivatives, of course, but when you're writing those out it's easier to see that you just wrote something like

  jacobians[i][j] = ... 1/(2*sqrt(value)) ...

and take steps to avoid those infinities.

-wjr

To unsubscribe from this group and stop receiving emails from it, send an email to ceres-solver+unsubscribe@googlegroups.com.

To view this discussion on the web visit https://groups.google.com/d/msgid/ceres-solver/10db297f-462c-46fb-bdbe-264737a3456a%40googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

--

You received this message because you are subscribed to the Google Groups "Ceres Solver" group.

To unsubscribe from this group and stop receiving emails from it, send an email to ceres-solver+unsubscribe@googlegroups.com.

To view this discussion on the web visit https://groups.google.com/d/msgid/ceres-solver/CABqdRUBmtdTMjrh-%3De8h-F4aaw2jF3BDu-pWPZ2jiqcOd5Yzcg%40mail.gmail.com.