NDEBUG

This warning was issued on OSX after calling

cmake -DBUILD_TESTING=OFF -DBUILD_EXAMPLES=ON -DPLOT_EXAMPLES=ON -DCMAKE_BUILD_TYPE=Release -DNDEBUG=1 -DCMAKE_CXX_FLAGS="-march=native -O3 -mtune=native -mavx2 -mfma" ..

If BUILD_TESTS = ON, I get this

CMake Error at test/CMakeLists.txt:35 (add_dependencies):
  The dependency target "gtest" of target "gtest_demo_se2" does not exist.
Call Stack (most recent call first):
  test/CMakeLists.txt:55 (kalmanif_add_gtest)


CMake Error at test/CMakeLists.txt:35 (add_dependencies):
  The dependency target "gtest" of target "gtest_demo_se3" does not exist.
Call Stack (most recent call first):
  test/CMakeLists.txt:56 (kalmanif_add_gtest)


CMake Error at test/CMakeLists.txt:35 (add_dependencies):
  The dependency target "gtest" of target "gtest_demo_se_2_3" does not exist.
Call Stack (most recent call first):
  test/CMakeLists.txt:57 (kalmanif_add_gtest)

If I switch them OFF, the errors disappear. However, I have no examples being compiled:

cd examples
ls

CMakeFiles  cmake_install.cmake  Makefile

Running demo_se3 for a while... seems to never end!

Maybe add a time line t =t_current / t_final, just to the right of the ------ separation line

Ran se3 with gnuplot installed and the correct build according to instructions. Where are the plots?

Dear, @joansola and @artivis

I'm looking into the implementation of the UKF on manifold from this repo.
I've just noticed, that in the case of the Right Invariance sigmas and increments are added to the current state using the left box plus operator. In which frame P (covariance) is expressed then? Local tangent space at X, or global at the group origin?
I expected that with Right Invariance increments and covariance are expressed in the local tangent space at X and thus right box plus operator shall be used.

At the same time, in the example from manif repo, the description matches the operator used to add sigma increments to the state.

*  The estimation error dx and its covariance P are expressed
*  in the tangent space at X.

https://github.com/artivis/manif/blob/805a0b2adf9435dd658fdad5606c342db4d889e8/examples/se2_localization_ukfm.cpp#L311

        // sigma points on manifold
        for (int i = 0; i < DoF; ++i)
        {
          s_j_p = X + manif::SE2Tangentd( xis.col(i));
          s_j_m = X + manif::SE2Tangentd(-xis.col(i));

          xis_new.col(i) = (X_new.lminus(s_j_p + u_est)).coeffs();
          xis_new.col(i + DoF) = (X_new.lminus(s_j_m + u_est)).coeffs();
        }

Thank you in advance for your answer!

Best regards,
Alex

Dear @artivis ,

I follow your code here:

I see so far that the result from the measurement model is expected to be a vector. Are there any thoughts regarding the support of measurement models which return group?
Particularly, when UEKF is used to fuse several high level sources of orientation expressed as SO3 members.

Here are my thoughts so far:

1. Use a container to store sigma points instead of columns in the matrix (std vector should be fine).
2. Use proper averaging and weighting methods to compute mean and covariance on manifold (average_biinvariant?).

I think I'll be able to contribute with a little help from your side.

Best regards,
Alex

First of all, I would to thank you for such a great contribution to the open-source community.

I am new to state estimation formulation using Lie Theory. I saw your paper describing the Lie Theory with such great examples and the repo manif.

I request to please add an example related to IEKF for IMU, and GPS fusion.

If you feel it is not relevant to add an example, I request to please share some resources by which I could implement IMU/GPS fusion using the library.

Dear, @artivis

I followed your discussion here regarding the need for composite/compound state support. I also see that in the manif repo, there is a way to do so with the Bundle<>. But there is no example in this repository of how to do that. Are there any impediments invisible to me that prevent using composite states?

The idea is to use manif/kalmanif for a practical application with a heterogeneous state:
SE(3) - robot pose().
SO(3), SE(3) - sensor locations, orientations.
1D vectors, or scalars - sensor biases.
(Optionally) 2D matrix - more complex sensor calibrations, homography, maybe members of other groups such as SIM(3).

Looking forward to your answer!

Best regards,
Alex

I have quickly checked out the se2 example. I wonder whether the obs models accept non-constant covariances, that is, something that allows the following update

ekf.update(gps_measurement_model, y_gps, R_gps);

Hi!

Could you please reference the paper used for the scaled unscented transform?
I assume that one: The Scaled Unscented Transformation Simon J. Julier, IDAK Industries, 901 Missouri Blvd., #179 Jefferson City, MO 65109.
But there are some differences in how weights are used.

Thank you for your answer!

Best regards,
Alex.

Hello doctor @artivis,
I learned invariant EKF has some advantages, e.g. consistent and guaranteed convergence etc, over traditional EKF recently in paper [The Invariant Extended Kalman filter as a stable observer, Axel Barrau, Silvere Bonnabel]. But the measurement model must is Left-invariant observations, zt = x * d or Right-invariant observations, zt = x.inv * d. I want to loosely couple imu with a lio(lidar-inertial-odometry) system, which can give SE(3) measurement. But it seems that lio measurement can not statisfy invariant EKF measurement, right ?
From your experience, In kalmanif lib , Square Root Extended Kalman Filter (SEKF) or Unscented Kalman Filter on manifolds (UKFM), which one should i to try in terms of accuracy and stability? Thanks for your help and time a lot!

Best regards
narutojxl