GTSAM [1] packs a lot of functionality for robotics projects, including Kalman filters [2]. A product of GTech (Frank Dellaert), a pro is there are many cool projects built on top of it [3], in C++ and Python. A con is GTSAM is not the cleanest, well-architected codebase (although far better than most products of academic labs).

[1] https://bitbucket.org/gtborg/gtsam

[2] https://github.com/borglab/gtsam/tree/develop/examples#kalma...

[3] e.g., https://github.com/gtrll/piper and https://github.com/gtrll/gpmp2

The one quite often used in Robotics + ROS combination is robot_localization (https://github.com/cra-ros-pkg/robot_localization). The library provides implementation for ekf and ukf and in my experience it's quite robust.

SsfPack [1] is a great implementation if you are willing to suffer through Ox.

[1] http://www.ssfpack.com/

Most of SsfPack is now implemented in python statsmodels: https://www.statsmodels.org/stable/statespace.html

Thanks for that! I was not aware of this.

SsfPack will still be faster though and be less memory hungry than Python. On the brief look I had, it also seems that the nonlinear/non-Gaussian simulation methods are not implemented.

I think I've used the OpenCV Kalman filter implementation at some point. Maybe that suits your needs

OpenCV only provides Linear Kalman Filter afaik. For nonlinear estimation you need to use Extended or Unscented (this one is often used in Drone).

I wrote estimation library in Go [1] last year which implements a lot of Kalman Filter alternatives and optimisations + smoothing

https://github.com/milosgajdos83/go-estimate